diff --git a/NER - Deep Learning techniques. RNN and Transformer Model.html b/NER - Deep Learning techniques. RNN and Transformer Model.html
new file mode 100644
index 000000000..a03c8f760
--- /dev/null
+++ b/NER - Deep Learning techniques. RNN and Transformer Model.html
@@ -0,0 +1,3842 @@
+
+
+
+
+
+
+
+
+
+NER - Deep Learning techniques. RNN and Transformer Model
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
NER - Deep Learning techniques. RNN and Transformer Model
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
1. Introduction
+
Named Entity Recognition (NER) is a crucial task in Natural Language Processing (NLP) that involves identifying and extracting named entities from unstructured text. NER presents challenges due to the inherent complexity and ambiguity of natural language. Still, it is essential for various NLP applications, including information retrieval, question answering, and machine translation. Deep learning techniques have demonstrated exceptional performance in NER tasks in recent years. Deep learning, a subfield of machine learning, uses artificial neural networks to learn from data and make predictions.
+
The Transformer model, introduced in 2017, has become a fundamental architecture in modern NLP and is widely used for NER tasks. It is known for its self-attention mechanism, which allows the model to weigh the importance of each word in the input sequence. Additionally, other deep learning architectures such as Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), and Bidirectional Encoder Representations from Transformers (BERT) have been successfully applied to NER.
+
In this literature review, we will explore and compare the effectiveness of RNNs and the Transformer model for Named Entity Recognition (NER) tasks.
+
+
1.1 Recurrent Neural Networks (RNN)
+
A Recurrent Neural Network (RNN) is a method of deep learning that uses sequential data or time-series data. They are often used in ordinal problems, where the output is not continues but it is discrete and ordered, and temporal problems, where data is collected over time and the order of the observations is important. These problems include speech recognition, language translation, and natural language processing (nlp). RNNs make use of training data by taking prior inputs and using them to influence current inputs and outputs. Utilizing prior inputs in this way is what creates the RNN’s memory. This is something that distinguishes RNNs from other deep learning methods. RNNs are also distinguished by the sharing of parameters across each layer in the network. A popular type of RNN architecture known as Long Short-Term Memory (LSTM) has been used in named entity recognition. LSTMs have individual units in the hidden layers of a neural network, each of which has three gates. These include an input gate that controls the input information that goes into a memory cell. A forget gate that controls the amount of historical information that passes through from the previous state, and an output gate that controls the amount of information that is passed on to the next step. RNNs allow information to persist across multiple steps which enable the network to capture dependencies and context. The architecture and abilities of RNNs allow it to be a useful tool in NER that has been proven to perform better than previous NER systems.
+
+
+
1.2 Transformer Model
+
The Transformer model is a groundbreaking neural network architecture introduced in the paper “Attention Is All You Need” (vaswani2017attention). The model is designed for sequence-to-sequence tasks, such as machine translation, and is known for its ability to process input sequences in parallel rather than sequentially. This parallel processing makes the Transformer model highly efficient and scalable. One of the key innovations of the Transformer model is the self-attention mechanism. Self-attention allows the model to weigh the importance of different words in the input sequence relative to each other when making predictions. The model uses multi-head attention, which means it can simultaneously attend to various input aspects. This ability to capture complex dependencies and relationships between words contributes to the model’s strong performance. The Transformer architecture consists of an encoder and decoder, each composed of multiple layers of self-attention and feed forward neural networks. The encoder processes the input sequence while the decoder generates the output sequence. The connections between the encoder and decoder are facilitated by attention mechanisms that allow the decoder to focus on different parts of the input sequence as it generates the output. Given its effectiveness and efficiency in handling sequence data, the Transformer model has become the foundation for many subsequent natural language processing (NLP) models and architectures.
+
+
+
1.3 Limitations
+
[In progress - Pending RNN Method]
+
Transformers are powerful neural network models commonly used in natural language processing tasks. However, they face two fundamental limitations: First, their large number of parameters results in high computational complexity, necessitating specialized hardware and substantial resources, especially for deep models with long input sequences. Second, transformers exhibit quadratic time complexity concerning input length due to the self-attention mechanism, which calculates attention scores between all token pairs. This quadratic complexity limits transformers’ ability to handle very long sequences efficiently. Researchers actively explore methods to mitigate these limitations and improve the scalability of transformers.(keles2022computational)
+
+
+
+
2 Methods
+
[In progress - Pending RNN Method]
+
+
2.1 RNN
+
+
+
2.2 Transformer Model
+
The Transformers model has a unique feature called the ‘attention mechanism.’ It helps the model make intelligent predictions by focusing on essential parts of the input. For example, when predicting the next word in ‘I have a pet ___,’ the attention mechanism considers the context (the word ‘pet’) to make a better guess, like ‘dog’ or ‘cat.’ It’s like the model knows which words deserve extra attention! ! Let’s look at the two most important algorithms for the attention mechanism, Scaled Dot-Product Attention formula, and the Single (Masked) Self- or Cross-Attention Head Formula (phuong2022formal).
Softmax (...): Function to normalize attention scores.
+
+
In this formula, the attention mechanism allows a model to “pay attention” to specific parts of input data. The attention scores are calculated using queries (Q) and keys (K). The scores are normalized using Softmax , producing attention weights. The output is a weighted sum of value vectors (V), where the weights represent the importance of each value for each query. This mechanism is popular in tasks like machine translation to focus on relevant words.
+
The Single (Masked) Self- or Cross-Attention Head Formula \(\begin{align*}\text{Attention}(Q, K, V) &= \text{softmax}\left(\frac{QK^\top + \text{Mask}}{\sqrt{d_k}}\right)V \\\end{align*}\)
+
+
Key Components:
+
+
Q: Query matrix (requests for information).
+
K: Key matrix (used to calculate relevance).
+
V: Value matrix (actual content to retrieve).
+
d_k: Dimension of the query/key vectors.
+
Mask: Matrix to prevent attention to specific tokens (e.g., future tokens).
+
Softmax (...): Function to normalize attention scores.
+
+
In this formula, the attention uses queries (Q) and keys (K) to calculate attention scores. An optional mask (Mask) is added to scores to prevent attending to specific tokens (e.g., in language modeling, future tokens are masked). Scores are normalized using Softmax, producing attention weights. The output is a weighted sum of value vectors (V), representing the importance of each value for each query. Masking is useful for sequence-to-sequence tasks and autoregressive models.
+
+
+
+
3 Data Analysis and Results
+
+
3.1 Dataset Description
+
I will use the dataset called corona2 from Kaggle to identify Natural Entity Recognition to identify Medical Condition, Medicine names and Pathogens. The dataset was manually tagged for training. I will use the Transformer model to apply the NER using python programming language. The dataset contains 31 observations and 4 attributes properly explained in the data definition.
+
Data Definition
+
+
+
+
+
+
+
+
+
Column Name
+
Type
+
Description
+
+
+
+
+
Text
+
string
+
Sentence including the labels
+
+
+
Starts
+
integer
+
Position on where the label starts
+
+
+
Ends
+
integer
+
Position on where the label ends
+
+
+
Labels
+
string
+
The label( Medical Condition, Medicine or Pathogen)
+
+
+
+
+
Labels:
+
+
Medical condition names (example: influenza, headache, malaria)
+
Medicine names (example : aspirin, penicillin, ribavirin, methotrexate)
+
Pathogens ( example: Corona Virus, Zika Virus, cynobacteria, E. Coli)
+
+
+
+
+
+
+
+
3.2 Data Preparation
+
The following python code will load the required libraries
import requests
+
+# Accessing the JSON file from GitHub
+url ='https://raw.githubusercontent.com/jsanc223/datasetCorona2/main/Corona2.json'
+# HTTP GET request to the raw URL
+response = requests.get(url)
+# I am checking if the request was successful (status code 200)
+if response.status_code ==200:
+# Parse the JSON data from the response
+ data = response.json()
+print('Success, the Json data was stored into the data variable')
+else:
+print('Failed to fetch JSON data:', response.status_code)
+
+
Success, the Json data was stored into the data variable
+
+
+
Parse json data into dictionary to manipulate data.
Convert data from Dictionary to Dataframe. I am only showing the Text and Labels for each sentence
+
+
import pandas as pd
+# I am initialing empty lists to store the data for the DataFrame
+texts = []
+starts = []
+ends = []
+labels = []
+
+# Iterate through the training_data to extract individual entity annotations
+for example in training_data:
+ text = example['text']
+for entity in example['entities']:
+ start, end, label = entity
+# Append data to the lists
+ texts.append(text)
+ starts.append(start)
+ ends.append(end)
+ labels.append(label)
+
+# Create a DataFrame from the lists
+df = pd.DataFrame({'text': texts, 'start': starts, 'end': ends, 'label': labels})
+df.head(5)
+
+
+
+
+
+
+
+
+
text
+
start
+
end
+
label
+
+
+
+
+
0
+
While bismuth compounds (Pepto-Bismol) decreas...
+
360
+
371
+
MEDICINE
+
+
+
1
+
While bismuth compounds (Pepto-Bismol) decreas...
+
383
+
408
+
MEDICINE
+
+
+
2
+
While bismuth compounds (Pepto-Bismol) decreas...
+
104
+
112
+
MEDICALCONDITION
+
+
+
3
+
While bismuth compounds (Pepto-Bismol) decreas...
+
679
+
689
+
MEDICINE
+
+
+
4
+
While bismuth compounds (Pepto-Bismol) decreas...
+
6
+
23
+
MEDICINE
+
+
+
+
+
+
+
+
+
+
3.3 Data statistics
+
+
Our analysis begins with examining the frequency of each label in the dataset. The dataset contains 134 instances of ‘Medical Condition’, 94 instances of ‘Medicine’, and 67 instances of ‘Pathogen’. This data provides an overview of label distribution.
+
+
+
import plotly.express as px
+
+# Count the occurrences of each label
+label_counts = df['label'].value_counts()
+
+# Create a DataFrame with labels and their respective counts
+df_counts = pd.DataFrame({'label': label_counts.index, 'count': label_counts.values})
+
+# Plot the frequency of each entity label using a bar plot in Plotly
+fig = px.bar(df_counts, x='label', y='count', text='count', color='label',
+ color_discrete_sequence=px.colors.qualitative.Plotly, title='Frequency of Entity Labels')
+
+# Display the counter label inside the bars
+fig.update_traces(textposition='inside')
+
+# Update axis titles
+fig.update_layout(xaxis_title='Entity Label', yaxis_title='Frequency')
+
+fig.show()
+
+
+
+
+
+
+
The chart below displays the distribution of biomedical Entity labels in the dataset. ‘Medical Condition’ is the most prevalent at 45.4%, followed by ‘Medicine’ at 31.9%, and ‘Pathogen’ at 22.7%. The chart provides insights into the dataset composition and label prevalence..
+
+
+
import plotly.express as px
+
+# Get the counts of each unique label
+label_counts = df['label'].value_counts()
+# Plot a pie chart using Plotly
+fig = px.pie(label_counts, values=label_counts.values, names=label_counts.index, title='Proportion of Entity Labels', hole=0.3)
+fig.update_traces(textinfo='percent+label', textfont_size=12)
+fig.show()
+
+
+
+
+
+
+
The histogram below visualizes the start positions of entity labels in the text. It reveals that most entities occur within the first five hundred words.
+
+
+
import plotly.express as px
+
+# Plot a histogram of entity start positions using Plotly
+fig = px.histogram(df, x='start', nbins=30, title='Histogram of Entity Start Positions')
+fig.update_layout(xaxis_title='Entity Start Position', yaxis_title='Frequency')
+fig.show()
+
+
+
+
+
+
+
The box plot below shows the start and end positions of entity labels, with the majority located below the five hundredth position.
+
+
+
import plotly.express as px
+
+# Create box plots for 'start' and 'end' columns using Plotly
+fig = px.box(df, y=['start', 'end'], points='all', title='Box Plots of Start and End Entity Positions')
+fig.update_layout(yaxis_title='Value', xaxis_title='Column')
+fig.show()
+
+
+
+
+
+
+
+
+
4 Transformer Model Results
+
[Pending]
+
+
+
5 Recurrent Neural Networks Results
+
[Pending]
+
+
+
6 Conclusion
+
[Pending]
+
+
+
7 References
+
[In Progress]
+
@misc{phuong2022formal,
+
title={Formal Algorithms for Transformers},
+
author={Mary Phuong and Marcus Hutter},
+
year={2022},
+
eprint={2207.09238},
+
archivePrefix={arXiv},
+
primaryClass={cs.LG}
+
}
+
@misc{keles2022computational,
+
title={On The Computational Complexity of Self-Attention},
+
author={Feyza Duman Keles and Pruthuvi Mahesakya Wijewardena and Chinmay Hegde},
+
year={2022},
+
eprint={2209.04881},
+
archivePrefix={arXiv},
+
primaryClass={cs.LG}
+
}
+
@misc{vaswani2017attention,
+
title={Attention Is All You Need},
+
author={Ashish Vaswani and Noam Shazeer and Niki Parmar and Jakob Uszkoreit and Llion Jones and Aidan N. Gomez and Lukasz Kaiser and Illia Polosukhin},
+
year={2017},
+
eprint={1706.03762},
+
archivePrefix={arXiv},
+
primaryClass={cs.CL}
+
}
+
+
+
+
+
+
+
+
+
+
\ No newline at end of file
diff --git a/NER - Deep Learning techniques. RNN and Transformer Model.qmd b/NER - Deep Learning techniques. RNN and Transformer Model.qmd
new file mode 100644
index 000000000..67e701d9e
--- /dev/null
+++ b/NER - Deep Learning techniques. RNN and Transformer Model.qmd
@@ -0,0 +1,304 @@
+---
+title: "NER - Deep Learning techniques. RNN and Transformer Model"
+format:
+ html:
+ embed-resources: true
+---
+
+# 1. Introduction
+
+Named Entity Recognition (NER) is a crucial task in Natural Language Processing (NLP) that involves identifying and extracting named entities from unstructured text. NER presents challenges due to the inherent complexity and ambiguity of natural language. Still, it is essential for various NLP applications, including information retrieval, question answering, and machine translation. Deep learning techniques have demonstrated exceptional performance in NER tasks in recent years. Deep learning, a subfield of machine learning, uses artificial neural networks to learn from data and make predictions.
+
+The Transformer model, introduced in 2017, has become a fundamental architecture in modern NLP and is widely used for NER tasks. It is known for its self-attention mechanism, which allows the model to weigh the importance of each word in the input sequence. Additionally, other deep learning architectures such as Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), and Bidirectional Encoder Representations from Transformers (BERT) have been successfully applied to NER.
+
+In this literature review, we will explore and compare the effectiveness of RNNs and the Transformer model for Named Entity Recognition (NER) tasks.
+
+## 1.1 Recurrent Neural Networks (RNN)
+
+A Recurrent Neural Network (RNN) is a method of deep learning that uses sequential data or time-series data. They are often used in ordinal problems, where the output is not continues but it is discrete and ordered, and temporal problems, where data is collected over time and the order of the observations is important. These problems include speech recognition, language translation, and natural language processing (nlp). RNNs make use of training data by taking prior inputs and using them to influence current inputs and outputs. Utilizing prior inputs in this way is what creates the RNN's memory. This is something that distinguishes RNNs from other deep learning methods. RNNs are also distinguished by the sharing of parameters across each layer in the network. A popular type of RNN architecture known as Long Short-Term Memory (LSTM) has been used in named entity recognition. LSTMs have individual units in the hidden layers of a neural network, each of which has three gates. These include an input gate that controls the input information that goes into a memory cell. A forget gate that controls the amount of historical information that passes through from the previous state, and an output gate that controls the amount of information that is passed on to the next step. RNNs allow information to persist across multiple steps which enable the network to capture dependencies and context. The architecture and abilities of RNNs allow it to be a useful tool in NER that has been proven to perform better than previous NER systems.
+
+## 1.2 Transformer Model
+
+The Transformer model is a groundbreaking neural network architecture introduced in the paper "Attention Is All You Need" (vaswani2017attention). The model is designed for sequence-to-sequence tasks, such as machine translation, and is known for its ability to process input sequences in parallel rather than sequentially. This parallel processing makes the Transformer model highly efficient and scalable. One of the key innovations of the Transformer model is the self-attention mechanism. Self-attention allows the model to weigh the importance of different words in the input sequence relative to each other when making predictions. The model uses multi-head attention, which means it can simultaneously attend to various input aspects. This ability to capture complex dependencies and relationships between words contributes to the model's strong performance. The Transformer architecture consists of an encoder and decoder, each composed of multiple layers of self-attention and feed forward neural networks. The encoder processes the input sequence while the decoder generates the output sequence. The connections between the encoder and decoder are facilitated by attention mechanisms that allow the decoder to focus on different parts of the input sequence as it generates the output. Given its effectiveness and efficiency in handling sequence data, the Transformer model has become the foundation for many subsequent natural language processing (NLP) models and architectures.
+
+## 1.3 Limitations
+
+***\[In progress - Pending RNN Method\]***
+
+Transformers are powerful neural network models commonly used in natural language processing tasks. However, they face two fundamental limitations: First, their large number of parameters results in high computational complexity, necessitating specialized hardware and substantial resources, especially for deep models with long input sequences. Second, transformers exhibit quadratic time complexity concerning input length due to the self-attention mechanism, which calculates attention scores between all token pairs. This quadratic complexity limits transformers' ability to handle very long sequences efficiently. Researchers actively explore methods to mitigate these limitations and improve the scalability of transformers.(keles2022computational)
+
+# **2 Methods**
+
+***\[In progress - Pending RNN Method\]***
+
+## 2.1 RNN
+
+## 2.2 Transformer Model
+
+The Transformers model has a unique feature called the 'attention mechanism.' It helps the model make intelligent predictions by focusing on essential parts of the input. For example, when predicting the next word in 'I have a pet \_\_\_,' the attention mechanism considers the context (the word 'pet') to make a better guess, like 'dog' or 'cat.' It's like the model knows which words deserve extra attention! ! Let's look at the two most important algorithms for the attention mechanism, Scaled Dot-Product Attention formula, and the Single (Masked) Self- or Cross-Attention Head Formula (phuong2022formal).
+
+1. Scaled Dot-Product Attention formula\
+ \
+ $\text{Attention}(Q, K, V) = \text{softmax}\left(\frac{QK^\top}{\sqrt{d_k}}\right)V$\
+
+ **Key Components:**
+
+ - Q: Query matrix (requests for information).
+
+ - K: Key matrix (used to calculate relevance).
+
+ - V: Value matrix (actual content to retrieve).
+
+ - d_k: Dimension of the query/key vectors.
+
+ - Softmax (\...): Function to normalize attention scores.
+
+ In this formula, the attention mechanism allows a model to "pay attention" to specific parts of input data. The attention scores are calculated using queries (Q) and keys (K). The scores are normalized using Softmax , producing attention weights. The output is a weighted sum of value vectors (V), where the weights represent the importance of each value for each query. This mechanism is popular in tasks like machine translation to focus on relevant words.
+
+2. The Single (Masked) Self- or Cross-Attention Head Formula $\begin{align*}\text{Attention}(Q, K, V) &= \text{softmax}\left(\frac{QK^\top + \text{Mask}}{\sqrt{d_k}}\right)V \\\end{align*}$
+
+**Key Components:**
+
+- Q: Query matrix (requests for information).
+
+- K: Key matrix (used to calculate relevance).
+
+- V: Value matrix (actual content to retrieve).
+
+- d_k: Dimension of the query/key vectors.
+
+- Mask: Matrix to prevent attention to specific tokens (e.g., future tokens).
+
+- Softmax (\...): Function to normalize attention scores.
+
+In this formula, the attention uses queries (Q) and keys (K) to calculate attention scores. An optional mask (Mask) is added to scores to prevent attending to specific tokens (e.g., in language modeling, future tokens are masked). Scores are normalized using Softmax, producing attention weights. The output is a weighted sum of value vectors (V), representing the importance of each value for each query. Masking is useful for sequence-to-sequence tasks and autoregressive models.
+
+# **3 Data Analysis and Results**
+
+## 3.1 Dataset Description
+
+I will use the dataset called corona2 from Kaggle to identify Natural Entity Recognition to identify Medical Condition, Medicine names and Pathogens. The dataset was manually tagged for training. I will use the Transformer model to apply the NER using python programming language. The dataset contains 31 observations and 4 attributes properly explained in the data definition.
+
+**Data Definition**
+
+| Column Name | Type | Description |
+|------------------|------------------|-----------------------------------|
+| Text | string | Sentence including the labels |
+| Starts | integer | Position on where the label starts |
+| Ends | integer | Position on where the label ends |
+| Labels | string | The label( Medical Condition, Medicine or Pathogen) |
+
+- Labels:
+ - Medical condition names (example: influenza, headache, malaria)
+ - Medicine names (example : aspirin, penicillin, ribavirin, methotrexate)
+ - Pathogens ( example: Corona Virus, Zika Virus, cynobacteria, E. Coli)
+
+##
+
+## 3.2 Data Preparation
+
+The following python code will load the required libraries
+
+```{python}
+!pip install -q nbclient
+!pip install -q requests
+!pip install -q pandas
+!pip install -q nbformat
+!pip install -q plotly.express
+```
+
+Import the data from Github to local.
+
+```{python}
+
+import requests
+
+# Accessing the JSON file from GitHub
+url = 'https://raw.githubusercontent.com/jsanc223/datasetCorona2/main/Corona2.json'
+# HTTP GET request to the raw URL
+response = requests.get(url)
+# I am checking if the request was successful (status code 200)
+if response.status_code == 200:
+ # Parse the JSON data from the response
+ data = response.json()
+ print('Success, the Json data was stored into the data variable')
+else:
+ print('Failed to fetch JSON data:', response.status_code)
+
+```
+
+Parse json data into dictionary to manipulate data.
+
+```{python}
+
+training_data = []
+for example in data['examples']:
+ temp_dict = {}
+ temp_dict['text'] = example['content']
+ temp_dict['entities'] = []
+ for annotation in example['annotations']:
+ start = annotation['start']
+ end = annotation['end']
+ label = annotation['tag_name'].upper()
+ temp_dict['entities'].append((start, end, label))
+ training_data.append(temp_dict)
+
+```
+
+Convert data from Dictionary to Dataframe. I am only showing the Text and Labels for each sentence
+
+```{python}
+import pandas as pd
+# I am initialing empty lists to store the data for the DataFrame
+texts = []
+starts = []
+ends = []
+labels = []
+
+# Iterate through the training_data to extract individual entity annotations
+for example in training_data:
+ text = example['text']
+ for entity in example['entities']:
+ start, end, label = entity
+ # Append data to the lists
+ texts.append(text)
+ starts.append(start)
+ ends.append(end)
+ labels.append(label)
+
+# Create a DataFrame from the lists
+df = pd.DataFrame({'text': texts, 'start': starts, 'end': ends, 'label': labels})
+df.head(5)
+```
+
+## 3.3 Data statistics
+
+1. Our analysis begins with examining the frequency of each label in the dataset. The dataset contains 134 instances of 'Medical Condition', 94 instances of 'Medicine', and 67 instances of 'Pathogen'. This data provides an overview of label distribution.
+
+```{python}
+
+import plotly.express as px
+
+# Count the occurrences of each label
+label_counts = df['label'].value_counts()
+
+# Create a DataFrame with labels and their respective counts
+df_counts = pd.DataFrame({'label': label_counts.index, 'count': label_counts.values})
+
+# Plot the frequency of each entity label using a bar plot in Plotly
+fig = px.bar(df_counts, x='label', y='count', text='count', color='label',
+ color_discrete_sequence=px.colors.qualitative.Plotly, title='Frequency of Entity Labels')
+
+# Display the counter label inside the bars
+fig.update_traces(textposition='inside')
+
+# Update axis titles
+fig.update_layout(xaxis_title='Entity Label', yaxis_title='Frequency')
+
+fig.show()
+
+```
+
+2. The chart below displays the distribution of biomedical Entity labels in the dataset. 'Medical Condition' is the most prevalent at 45.4%, followed by 'Medicine' at 31.9%, and 'Pathogen' at 22.7%. The chart provides insights into the dataset composition and label prevalence..
+
+```{python}
+
+import plotly.express as px
+
+# Get the counts of each unique label
+label_counts = df['label'].value_counts()
+# Plot a pie chart using Plotly
+fig = px.pie(label_counts, values=label_counts.values, names=label_counts.index, title='Proportion of Entity Labels', hole=0.3)
+fig.update_traces(textinfo='percent+label', textfont_size=12)
+fig.show()
+
+```
+
+3. The histogram below visualizes the start positions of entity labels in the text. It reveals that most entities occur within the first five hundred words.
+
+```{python}
+
+import plotly.express as px
+
+# Plot a histogram of entity start positions using Plotly
+fig = px.histogram(df, x='start', nbins=30, title='Histogram of Entity Start Positions')
+fig.update_layout(xaxis_title='Entity Start Position', yaxis_title='Frequency')
+fig.show()
+```
+
+4. The box plot below shows the start and end positions of entity labels, with the majority located below the five hundredth position.
+
+```{python}
+
+import plotly.express as px
+
+# Create box plots for 'start' and 'end' columns using Plotly
+fig = px.box(df, y=['start', 'end'], points='all', title='Box Plots of Start and End Entity Positions')
+fig.update_layout(yaxis_title='Value', xaxis_title='Column')
+fig.show()
+
+```
+
+# **4 Transformer Model Results**
+
+***\[Pending\]***
+
+# **5 Recurrent Neural Networks Results**
+
+***\[Pending\]***
+
+# **6 Conclusion**
+
+***\[Pending\]***
+
+# 7 References
+
+\[In Progress\]
+
+\@misc{phuong2022formal,
+
+title={Formal Algorithms for Transformers},
+
+author={Mary Phuong and Marcus Hutter},
+
+year={2022},
+
+eprint={2207.09238},
+
+archivePrefix={arXiv},
+
+primaryClass={cs.LG}
+
+}
+
+\@misc{keles2022computational,
+
+title={On The Computational Complexity of Self-Attention},
+
+author={Feyza Duman Keles and Pruthuvi Mahesakya Wijewardena and Chinmay Hegde},
+
+year={2022},
+
+eprint={2209.04881},
+
+archivePrefix={arXiv},
+
+primaryClass={cs.LG}
+
+}
+
+\@misc{vaswani2017attention,
+
+title={Attention Is All You Need},
+
+author={Ashish Vaswani and Noam Shazeer and Niki Parmar and Jakob Uszkoreit and Llion Jones and Aidan N. Gomez and Lukasz Kaiser and Illia Polosukhin},
+
+year={2017},
+
+eprint={1706.03762},
+
+archivePrefix={arXiv},
+
+primaryClass={cs.CL}
+
+}
diff --git a/README.md b/README.md
index 15749d71a..63eeac51b 100644
--- a/README.md
+++ b/README.md
@@ -1,21 +1,7 @@
-# Instructions
+# Name Entity Recognition (NER) using Deep Learning Techniques
-Create a GitHub page for the project
+## Group Members:
+#### 1. Sara Ericson
+#### 2. Andrew Campbell
+#### 2. Jorge Sanchez
-## GitHub:
-#### 1. Create a GitHub account and Sign in
-#### 2. Go to https://github.com/acohenstat/STA6257_Project and fork (create a copy to your GitHub)
-
-#### 3. Change the name of the repo to *STA6257_Project_NameofGroup*
-#### 4. Go to *Settings* -> *Pages* -> under *Branch* -> select *main*
-#### 5. Wait for a few seconds and refresh the page. You see the link of the page.
-
-## RStudio:
-#### 1. Go to RStudio
-#### 2. Create a Version Control Project and Clone the repo.
-#### 3. Commit and push to see changes on the website et Voilà!
-
-More information:
-- [GitHub](https://happygitwithr.com/index.html)
-- [Video1 RStudio connection to GitHub](https://www.youtube.com/watch?v=MdmnE3AnkQE)
-- [Video2 RStudio connection to GitHub](https://www.youtube.com/watch?v=jN6tvgt3GK8)
diff --git a/images/annotated1.png b/images/annotated1.png
new file mode 100644
index 000000000..42c206888
Binary files /dev/null and b/images/annotated1.png differ
diff --git a/images/annotated2.png b/images/annotated2.png
new file mode 100644
index 000000000..36d7c9a49
Binary files /dev/null and b/images/annotated2.png differ
diff --git a/images/frequency_labels-01.png b/images/frequency_labels-01.png
new file mode 100644
index 000000000..328ed4366
Binary files /dev/null and b/images/frequency_labels-01.png differ
diff --git a/images/start_end_Positions-01.png b/images/start_end_Positions-01.png
new file mode 100644
index 000000000..3ab80d8c0
Binary files /dev/null and b/images/start_end_Positions-01.png differ
diff --git a/images/transformer Architecture.png b/images/transformer Architecture.png
new file mode 100644
index 000000000..2cffd5599
Binary files /dev/null and b/images/transformer Architecture.png differ
diff --git a/index.html b/index.html
index f8ab42114..96924dd20 100644
--- a/index.html
+++ b/index.html
@@ -2,21 +2,27 @@
-
+
-
-
+
+
-Report Sample
+NER using RNNs and Transformer Models
+}
@@ -111,8 +87,9 @@
-
+
+
@@ -125,7 +102,7 @@
-
Report Sample
+
NER using RNNs and Transformer Models
@@ -135,159 +112,694 @@
Report Sample
Author
-
Student name
+
Sara Ericson, Andrew Campbell, Jorge Sanchez
Published
-
October 25, 2022
+
April 25, 2023
+
-
-
Introduction
-
Kernel regression is a non-parametric estimator that estimates the conditional expectation of two variables which is random. The goal of a kernel regression is to discover the non-linear relationship between two random variables. To discover the non-linear relationship, kernel estimator or kernel smoothing is the main method to estimate the curve for non-parametric statistics. In kernel estimator, weight function is known as kernel function (Efromovich 2008). Cite this paper (Bro and Smilde 2014).
The common non-parametric regression model is \(Y_i = m(X_i) + \varepsilon_i\), where \(Y_i\) can be defined as the sum of the regression function value \(m(x)\) for \(X_i\). Here \(m(x)\) is unknown and \(\varepsilon_i\) some errors. With the help of this definition, we can create the estimation for local averaging i.e. \(m(x)\) can be estimated with the product of \(Y_i\) average and \(X_i\) is near to \(x\). In other words, this means that we are discovering the line through the data points with the help of surrounding data points. The estimation formula is printed below (R Core Team 2019):
\(W_n(x)\) is the sum of weights that belongs to all real numbers. Weights are positive numbers and small if \(X_i\) is far from \(x\).
+
+
1. Introduction
+
Named Entity Recognition (NER) is a crucial task in Natural Language Processing (NLP) that involves identifying and extracting named entities from unstructured text. NER presents challenges due to the inherent complexity and ambiguity of natural language. Still, it is essential for various NLP applications, including information retrieval, question answering, and machine translation. Biomedical NER is the task of text mining to specifically biomedical texts to determine entity types (xiong2021Improving?). Similarly, medical NER tasks usually find the boundaries of mentions from medical texts (Zhao_Liu_Zhao_Wang_2019?). Deep learning techniques have demonstrated exceptional performance in NER tasks in recent years. Deep learning, a subfield of machine learning, uses artificial neural networks to learn from data and make predictions.
+
The Transformer model, introduced in 2017, has become a fundamental architecture in modern NLP and is widely used for NER tasks. It is known for its self-attention mechanism, which allows the model to weigh the importance of each word in the input sequence. Additionally, other deep learning architectures such as Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), and Bidirectional Encoder Representations from Transformers (BERT) have been successfully applied to NER.
+
In this literature review, we will explore and compare the effectiveness of RNNs and the Transformer model for Named Entity Recognition (NER) tasks.
+
+
1.1 Recurrent Neural Networks (RNN)
+
A Recurrent Neural Network (RNN) is a method of deep learning that is useful for sequential data and time-series data. RNNs are often used in ordinal problems, where the output is discrete and ordered, as well as temporal problems, in which the time that data is collected is important. These problems include speech recognition, language translation, and natural language processing (NLP). RNNs are similar to other neural networks, but they have some distinct differences. Other neural network models can be used for medical text mining, however, RNNs have been shown to be one of the most effective models. This model has some drawbacks, but they can be mitigated with the use of Long Short-Term Memory (LSTM) cells in the hidden layers of the neural network (Wu_Jiang_Xu_Zhi_Xu2018ClinicalNER?). RNNs have the ability to allow information to persist across multiple steps which enables the network to capture dependencies and context. This is a feature that makes it ideal for NER in which the context is very important. The specifics of a recurrent neural network will be discussed later
The Transformer model is a groundbreaking neural network architecture introduced in the paper “Attention Is All You Need” (Vaswani et al. 2017). The model is designed for sequence-to-sequence tasks, such as machine translation, and is known for its ability to process input sequences in parallel rather than sequentially. This parallel processing makes the Transformer model highly efficient and scalable. One of the key innovations of the Transformer model is the self-attention mechanism. Self-attention allows the model to weigh the importance of different words in the input sequence relative to each other when making predictions. The model uses multi-head attention, which means it can simultaneously attend to various input aspects. This ability to capture complex dependencies and relationships between words contributes to the model’s strong performance. The Transformer architecture consists of an encoder and decoder, each composed of multiple layers of self-attention and feed forward neural networks. The encoder processes the input sequence while the decoder generates the output sequence. The connections between the encoder and decoder are facilitated by attention mechanisms that allow the decoder to focus on different parts of the input sequence as it generates the output. Given its effectiveness and efficiency in handling sequence data, the Transformer model has become the foundation for many subsequent natural language processing (NLP) models and architectures.
+
+
+
1.3 Limitations
+
A major limitation of current techniques is that the standard language models are unidirectional, and thus limit the choice of architecture that can be used for pre-training (devlin2019bert?).
+
Transformers are powerful neural network models commonly used in natural language processing tasks. However, they face two fundamental limitations: First, their large number of parameters results in high computational complexity, necessitating specialized hardware and substantial resources, especially for deep models with long input sequences. Second, transformers exhibit quadratic time complexity concerning input length due to the self-attention mechanism, which calculates attention scores between all token pairs. This quadratic complexity limits transformers’ ability to handle very long sequences efficiently. Researchers actively explore methods to mitigate these limitations and improve the scalability of transformers.(Keles, Wijewardena, and Hegde 2022)
+
+
+
+
2 Methods
+
+
2.1 RNN
+
A neural network has three layers. An input layer, a hidden layer, and an output layer. A recurrent neural network (RNN) also has three layers, but with the addition of a recurrent component that takes the output from a previous iteration and includes it in the current iteration. This is what makes RNNs useful for sequential and temporal data. Basic neural networks are optimized using back propagation which involves finding the gradient of a parameter and then using a gradient descent algorithm to find values that minimize a loss function (such as error).
+
Example of Loss function (Mean Squared Error)
+
+\(\text{L(𝜽)} = (1/N) * ∑(y_i - y_i^*)^2\)
+
N : number of vector entires with yi in output vector
+
yi : predicted value
+
yi* : actual value
+
𝛼 : learning rate
+
𝜽j : input
+
This will cause the weight parameters to be updated. When a large amount of data is used this can cause the weights to increase or decrease dramatically creating a problem known as vanishing or exploding gradients. When this occurs the weights will either be extremely close to zero or extremely large, which may be represented by a NaN value. An example when large amounts of data may be used is NER in which the context of previous words and characters is relevant. To mitigate an exploding or vanishing gradient long short-term memory (LSTM) cells may be used within the hidden layer. LSTM cells make use of three gates. An input gate, a forget gate, and an output gate. These gates use a sigmoid function which outputs a value between zero and one. Zero blocks all information and one allows all information through. The input gate determines which information is stored in the cell. The forget gate determines which information will be discarded. The output gate provides the activation for the final output. The sigmoid function is given below along with a general equation for the gates. When using LSTMs the weights and biases are the same among every iteration. This is so the RNN model can be applied to data of varying lengths.(Cho2019BiomedNER?)
LSTMs have a cell state that is the result of the gates and a few other calculations. This state is what provides the cell with its “memory” which is very useful in NER when the memory of what was written or previously said can determine what should come next in a sentence. The other equations used besides sigmoid are shown below.
+
Possible cell state at i-th iteration
+
+\(\text{c}_i^*=tanh(w[h_{i-1}, x_i] + b)\)
+
These equations determine which information is passed through the cell and the ultimate state of the cell at the present iteration. The type of RNN used in NER is many-to-many RNN. This type of RNN uses each word as an input to build the state (the context) of the model. This is what can be subsequently used in making predictions.
+
+
+
2.2 Transformer Model
+
The Transformers model has a unique feature called the ‘attention mechanism.’ It helps the model make intelligent predictions by focusing on essential parts of the input. For example, when predicting the next word in ‘I have a pet ___,’ the attention mechanism considers the context (the word ‘pet’) to make a better guess, like ‘dog’ or ‘cat.’ It’s like the model knows which words deserve extra attention! ! Let’s look at the two most important algorithms for the attention mechanism, Scaled Dot-Product Attention formula, and the Single (Masked) Self- or Cross-Attention Head Formula (Phuong and Hutter 2022).
Softmax (...): Function to normalize attention scores.
+
+
In this formula, the attention mechanism allows a model to “pay attention” to specific parts of input data. The attention scores are calculated using queries (Q) and keys (K). The scores are normalized using Softmax , producing attention weights. The output is a weighted sum of value vectors (V), where the weights represent the importance of each value for each query. This mechanism is popular in tasks like machine translation to focus on relevant words.
+
The Single (Masked) Self- or Cross-Attention Head Formula \(\begin{align*}\text{Attention}(Q, K, V) &= \text{softmax}\left(\frac{QK^\top + \text{Mask}}{\sqrt{d_k}}\right)V \\\end{align*}\)
+
+
Key Components:
+
+
Q: Query matrix (requests for information).
+
K: Key matrix (used to calculate relevance).
+
V: Value matrix (actual content to retrieve).
+
d_k: Dimension of the query/key vectors.
+
Mask: Matrix to prevent attention to specific tokens (e.g., future tokens).
+
Softmax (...): Function to normalize attention scores.
+
+
In this formula, the attention uses queries (Q) and keys (K) to calculate attention scores. An optional mask (Mask) is added to scores to prevent attending to specific tokens (e.g., in language modeling, future tokens are masked). Scores are normalized using Softmax, producing attention weights. The output is a weighted sum of value vectors (V), representing the importance of each value for each query. Masking is useful for sequence-to-sequence tasks and autoregressive models.
+
+
+
+
3 Data Analysis and Results
+
+
3.1 Dataset Description
+
We will use the dataset called corona2 from Kaggle to identify Natural Entity Recognition to identify Medical Condition, Medicine names and Pathogens. Similarly to the dataset used in (xiong2021Improving?), the dataset was manually tagged for training. We will use the Transformer model and the RNN model to apply the NER using python programming language. The dataset contains 31 observations and 4 attributes properly explained in the data definition.
+
Data Definition
+
+
+
+
+
+
-
state
-
abb
-
region
-
population
-
total
+
Column Name
+
Type
+
Description
-
Alabama
-
AL
-
South
-
4779736
-
135
-
-
-
Alaska
-
AK
-
West
-
710231
-
19
-
-
-
Arizona
-
AZ
-
West
-
6392017
-
232
+
Text
+
string
+
Sentence including the labels
-
Arkansas
-
AR
-
South
-
2915918
-
93
+
Starts
+
integer
+
Position on where the label starts
-
California
-
CA
-
West
-
37253956
-
1257
+
Ends
+
integer
+
Position on where the label ends
-
Colorado
-
CO
-
West
-
5029196
-
65
+
Labels
+
string
+
The label( Medical Condition, Medicine or Pathogen)
+
+
Labels:
+
+
Medical condition names (example: influenza, headache, malaria)
+
Medicine names (example : aspirin, penicillin, ribavirin, methotrexate)
+
Pathogens ( example: Corona Virus, Zika Virus, cynobacteria, E. Coli)
+
+
+
+
+
+
+
+
3.2 Data Preparation
+
The following python code will load the required libraries:
url ='https://raw.githubusercontent.com/jsanc223/datasetCorona2/main/Corona2.json'
+# HTTP GET request to the raw URL
+response = requests.get(url)
+# I am checking if the request was successful (status code 200)
+if response.status_code ==200:
+# Parse the JSON data from the response
+ data = response.json()
+print('Success, the Json data was stored into the data variable')
+else:
+print('Failed to fetch JSON data:', response.status_code)
+
+
+
Success, the Json data was stored into the data variable
+
+
+
Parse json data into dictionary to manipulate data.
Convert data from Dictionary to Dataframe. I am only showing the Text and Labels for each sentence
+
+
+Code
+
import pandas as pd
+# I am initialing empty lists to store the data for the DataFrame
+texts = []
+starts = []
+ends = []
+labels = []
+
+# Iterate through the training_data to extract individual entity annotations
+for example in training_data:
+ text = example['text']
+for entity in example['entities']:
+ start, end, label = entity
+# Append data to the lists
+ texts.append(text)
+ starts.append(start)
+ ends.append(end)
+ labels.append(label)
+
+# Create a DataFrame from the lists
+df = pd.DataFrame({'text': texts, 'start': starts, 'end': ends, 'label': labels})
+df.head(5)
+
+
+
text ... label
+0 While bismuth compounds (Pepto-Bismol) decreas... ... MEDICINE
+1 While bismuth compounds (Pepto-Bismol) decreas... ... MEDICINE
+2 While bismuth compounds (Pepto-Bismol) decreas... ... MEDICALCONDITION
+3 While bismuth compounds (Pepto-Bismol) decreas... ... MEDICINE
+4 While bismuth compounds (Pepto-Bismol) decreas... ... MEDICINE
+
+[5 rows x 4 columns]
+
+
+
+
+
3.3 Data statistics
+
+
Our analysis begins with examining the frequency of each label in the dataset. The dataset contains 134 instances of ‘Medical Condition’, 94 instances of ‘Medicine’, and 67 instances of ‘Pathogen’. This data provides an overview of label distribution.
+
+
+
+Code
+
+import plotly.express as px
+
+# Count the occurrences of each label
+label_counts = df['label'].value_counts()
+
+# Create a DataFrame with labels and their respective counts
+df_counts = pd.DataFrame({'label': label_counts.index, 'count': label_counts.values})
+
+# Plot the frequency of each entity label using a bar plot in Plotly
+fig = px.bar(df_counts, x='label', y='count', text='count', color='label',
+ color_discrete_sequence=px.colors.qualitative.Plotly, title='Frequency of Entity Labels')
+
+fig.update_layout(xaxis_title='Entity Label', yaxis_title='Frequency')
+
+# Display the counter label inside the bars
+#fig.update_traces(textposition='inside')
+# Update axis titles
+#fig.update_layout(xaxis_title='Entity Label', yaxis_title='Frequency')
+
+#fig.show()
-
+
+
+
+
+
+
+
The chart below displays the distribution of biomedical Entity labels in the dataset. ‘Medical Condition’ is the most prevalent at 45.4%, followed by ‘Medicine’ at 31.9%, and ‘Pathogen’ at 22.7%. The chart provides insights into the dataset composition and label prevalence..
+
+
+
+Code
+
+import plotly.express as px
+
+# Get the counts of each unique label
+label_counts = df['label'].value_counts()
+# Plot a pie chart using Plotly
+fig = px.pie(label_counts, values=label_counts.values, names=label_counts.index, title='Proportion of Entity Labels', hole=0.3)
+fig.update_traces(textinfo='percent+label', textfont_size=12)
+#fig.show()
+
+
+
+
+
+
+
+
+
The histogram below visualizes the start positions of entity labels in the text. It reveals that most entities occur within the first five hundred words.
+
+
+
+Code
+
+import plotly.express as px
+
+# Plot a histogram of entity start positions using Plotly
+fig = px.histogram(df, x='start', nbins=30, title='Histogram of Entity Start Positions')
+fig.update_layout(xaxis_title='Entity Start Position', yaxis_title='Frequency')
+#fig.show()
+
+
+
+
+
+
+
+
+
The box plot below shows the start and end positions of entity labels, with the majority located below the five hundredth position.
+
+
+
+Code
+
+import plotly.express as px
+
+# Create box plots for 'start' and 'end' columns using Plotly
+fig = px.box(df, y=['start', 'end'], points='all', title='Box Plots of Start and End Entity Positions')
+fig.update_layout(yaxis_title='Value', xaxis_title='Column')
+#fig.show()
+
+
+
+
+
-
-
Statistical Modeling
-
-
Conlusion
+
+
4 Transformer Model Results
+
[Pending]
+
Loading Required Libraries for training
+
+
+Code
+
#!pip install transformers
+#!pip install torch
+
+
+
The following code will retrieve the hosted Json dataset from my Github repositoy
+
+
+Code
+
import requests
+
+# Extracting the raw Json data from GitHub URL. I am hosting the dataset in my own repository
+url ='https://raw.githubusercontent.com/jsanc223/datasetCorona2/main/Corona2.json'
+# HTTP GET request to the raw URL
+response = requests.get(url)
+# Check if the request was successful (status code 200)
+if response.status_code ==200:
+# Parse the JSON data from the response
+ data = response.json()
+print('We retrieved the Json data successfully')
+else:
+print('Failed to fetch JSON data:', response.status_code)
+
+
+
We retrieved the Json data successfully
+
+
+
The following code block formats the raw data into a structure suitable for training a NER model. The raw data contains multiple examples, each with text content and annotations. The annotations include the start and end positions of entities and their corresponding labels. The output is a list of dictionaries, each representing an example containing the text and a list of entities with their positions and labels.
+
+
+Code
+
training_data = []
+for example in data['examples']:
+ temp_dict = {}
+ temp_dict['text'] = example['content']
+ temp_dict['entities'] = []
+for annotation in example['annotations']:
+ start = annotation['start']
+ end = annotation['end']
+ label = annotation['tag_name'].upper()
+ temp_dict['entities'].append((start, end, label))
+ training_data.append(temp_dict)
+print('Our dataset contains the following number of samples: ', len(training_data))
+
+
+
Our dataset contains the following number of samples: 31
+
+
+
This code formats entity annotations for our NER task. It selects the first training example, retrieves the text and annotated entities, and presents them in a table with start and end positions and labels. The output includes the text and a table of entities.
+
+
+Code
+
import pandas as pd
+example = training_data[0]
+
+# Extract text and entities from the selected example
+text = example['text']
+entities = example['entities']
+
+# Create a dataframe to store the entities with column names: Start, End, Label
+entities_df = pd.DataFrame(entities, columns=['Start', 'End', 'Label'])
+
+# Display the text and entities table
+print("Text:", text)
+
+
+
Text: While bismuth compounds (Pepto-Bismol) decreased the number of bowel movements in those with travelers' diarrhea, they do not decrease the length of illness.[91] Anti-motility agents like loperamide are also effective at reducing the number of stools but not the duration of disease.[8] These agents should be used only if bloody diarrhea is not present.[92]
+
+Diosmectite, a natural aluminomagnesium silicate clay, is effective in alleviating symptoms of acute diarrhea in children,[93] and also has some effects in chronic functional diarrhea, radiation-induced diarrhea, and chemotherapy-induced diarrhea.[45] Another absorbent agent used for the treatment of mild diarrhea is kaopectate.
+
+Racecadotril an antisecretory medication may be used to treat diarrhea in children and adults.[86] It has better tolerability than loperamide, as it causes less constipation and flatulence.[94]
+
+
+Code
+
print("Entities:")
+
+
+
Entities:
+
+
+Code
+
print(entities_df)
+
+
+
Start End Label
+0 360 371 MEDICINE
+1 383 408 MEDICINE
+2 104 112 MEDICALCONDITION
+3 679 689 MEDICINE
+4 6 23 MEDICINE
+5 25 37 MEDICINE
+6 461 470 MEDICALCONDITION
+7 577 589 MEDICINE
+8 853 865 MEDICALCONDITION
+9 188 198 MEDICINE
+10 754 762 MEDICALCONDITION
+11 870 880 MEDICALCONDITION
+12 823 833 MEDICINE
+13 852 853 MEDICALCONDITION
+14 461 469 MEDICALCONDITION
+15 535 543 MEDICALCONDITION
+16 692 704 MEDICINE
+17 563 571 MEDICALCONDITION
+
+
+
The following code constructs a DataFrame that consolidates entity annotations from the training data for our NER task. The code extracts the text, start position, end position, and label for each annotated entity in the training examples. The resulting DataFrame organizes this information in columns, providing a comprehensive view of all entity annotations across the training dataset.
+
+
+Code
+
import pandas as pd
+# Initialize empty lists to store the data for the DataFrame
+texts = []
+starts = []
+ends = []
+labels = []
+
+# Iterate through the training_data to extract individual entity annotations
+for example in training_data:
+ text = example['text']
+for entity in example['entities']:
+ start, end, label = entity
+# Append data to the lists
+ texts.append(text)
+ starts.append(start)
+ ends.append(end)
+ labels.append(label)
+
+# Create a DataFrame from the lists
+df = pd.DataFrame({'text': texts, 'start': starts, 'end': ends, 'label': labels})
+df
+
+
+
text ... label
+0 While bismuth compounds (Pepto-Bismol) decreas... ... MEDICINE
+1 While bismuth compounds (Pepto-Bismol) decreas... ... MEDICINE
+2 While bismuth compounds (Pepto-Bismol) decreas... ... MEDICALCONDITION
+3 While bismuth compounds (Pepto-Bismol) decreas... ... MEDICINE
+4 While bismuth compounds (Pepto-Bismol) decreas... ... MEDICINE
+.. ... ... ...
+290 Influenza, commonly known as "the flu", is an ... ... MEDICALCONDITION
+291 Influenza, commonly known as "the flu", is an ... ... MEDICALCONDITION
+292 Influenza, commonly known as "the flu", is an ... ... MEDICALCONDITION
+293 Influenza, commonly known as "the flu", is an ... ... MEDICALCONDITION
+294 Influenza, commonly known as "the flu", is an ... ... PATHOGEN
+
+[295 rows x 4 columns]
+
+
+
The Transformer model needed to be deployed in a Google Colab Repo because our computers did not have the power to train the transformer model when running out of a CPU. We get the following results which are good besides the eval_loss with over 31%. The hyperparameters were modified by increasing the epochs from 3 to 5 and 5 to 10, but the loss decreased to 25%. Precision and accuracy did not improve. The best approach to reduce the loss is to get more data. We could not find the more annotated dataset for NER in health. In Google Colab, a Hardware accelerator (GPU) was used to increase the training speed, and it worked. The link for further review is provided Google Colab Code
Using the dataframe created from the transformer model a vocabulary of words and labels was created and then converted into indices. The sequences are then padded to a maximum length and then converted to one-hot encoded vectors. A one-hot encoded vector is a binary vector in which the label is encoded as 1 and everything else that is not the label is encoded as 0. This is necessary to train the model with tensorflow and recurrent neural networks.
+
+
+Code
+
words =set(df['text'].values)
+word2idx = {w: i +2for i, w inenumerate(words)}
+word2idx['PAD'] =0
+word2idx['UNK'] =1
+
+tags =set(df['label'].values)
+tag2idx = {t: i +1for i, t inenumerate(tags)}
+tag2idx['PAD'] =0
+
+X = [[word2idx.get(w, 1) for w in sentence.split()] for sentence in df['text'].values]
+y = [[tag2idx[t] for t in sentence.split()] for sentence in df['label'].values]
+
+maxlen =max(len(x) for x in X)
+X = pad_sequences(X, padding='post', maxlen=maxlen)
+y = pad_sequences(y, padding='post', maxlen=maxlen)
+y = to_categorical(y, num_classes=len(tag2idx))
+
+
+
After the data is manipulated the RNN model is created and compiled with the use of tensorflow.
Here the RNN model is shown to be very effective with an accuracy of 0.9967.
-
-
+
+
6 Conclusion
+
[This is a working conclusion]
+
Named Entity Recognition (NER) is a key NLP task that involves identifying named entities in text. With advances in deep learning and large availability of computational resources, methods based on deep learning have demonstrated obvious advantages over traditional methods and have become mainstream in the NER field (liu2021Hybrid?). Deep learning techniques, such as Recurrent Neural Networks (RNNs) and the Transformer Model, are clearly effective in NER Tasks. RNNs, including LSTMs, excel at processing sequences and capturing context, while the Transformer model is known for its self-attention mechanism and efficient parallel processing. Both architectures contribute to advancing NER, and the choice between them depends on the specific NLP problem. In the NLP problem in this paper it can be seen that the RNN model has a higher accuracy than the transformer model, so the RNN model would be the preferred method to use in NER practices. This conclusion coicides with (yadav2019survey?), who agreed that neural network models generally outperform feature-engineered models. As deep learning research progresses, we anticipate further advancements in NER performance using these techniques.
-Bro, Rasmus, and Age K Smilde. 2014. “Principal Component Analysis.”Analytical Methods 6 (9): 2812–31.
+
References
+
+Keles, Feyza Duman, Pruthuvi Mahesakya Wijewardena, and Chinmay Hegde. 2022. “On the Computational Complexity of Self-Attention.”https://arxiv.org/abs/2209.04881.
-
-Efromovich, S. 2008. Nonparametric Curve Estimation: Methods, Theory, and Applications. Springer Series in Statistics. Springer New York. https://books.google.com/books?id=mdoLBwAAQBAJ.
+
-R Core Team. 2019. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing. https://www.R-project.org.
+
+Vaswani, Ashish, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. “Attention Is All You Need.”https://arxiv.org/abs/1706.03762.
The Single (Masked) Self- or Cross-Attention Head Formula \(\begin{align*}\text{Attention}(Q, K, V) &= \text{softmax}\left(\frac{QK^\top + \text{Mask}}{\sqrt{d_k}}\right)V \\\end{align*}\)
+
+
+
+
Transformer - How to do it?
+
+
Annotated Dataset for Training
+
+
Labelstud.io , Prodi.gy
+
+
Pre-process the dataset
+
+
Tokenization - NLTK
+
+
Fine-tune a pre-trained Tranformer Model
+
+
Huggingface - BERT
+
+
Train
+
Evaluate
+
+
Metrics: Precision,recall and F1 score.
+
+
+
+
+
Transformer Limitations
+
+
Standard language models are unidirectional, restricting pre-training architecture options and limiting context awareness.
+
Transformers have high computational complexity due to numerous parameters, requiring significant resources and specialized hardware for deep models and long sequences.
+
+
+
+
Dataset
+
We will use the dataset called corona2 from Kaggle to identify Natural Entity Recognition to identify Medical Condition, Medicine names and Pathogens. This dataset is already annotated.
+
+
Labels:
+
+
Medical condition names (example: influenza, headache, malaria)
+
Medicine names (example : aspirin, penicillin, ribavirin, methotrexate)
+
Pathogens ( example: Corona Virus, Zika Virus, cynobacteria, E. Coli)
+
+
+
+
+
Dataset Definition
+
+
+
+
+
+
+
+
+
Column Name
+
Type
+
Description
+
+
+
+
+
Text
+
string
+
Sentence including the labels
+
+
+
Starts
+
integer
+
Position on where the label starts
+
+
+
Ends
+
integer
+
Position on where the label ends
+
+
+
Labels
+
string
+
The label( Medical Condition, Medicine or Pathogen)
+
+
+
+
+
+
Dataset Sample
+
Buprenorphine has been shown experimentally (1982–1995) to be effective against severe, refractory depression.
\n\t\t\t`,K.overlay.querySelector(".close").addEventListener("click",(e=>{ze(),e.preventDefault()}),!1)}}function ze(){return!!K.overlay&&(K.overlay.parentNode.removeChild(K.overlay),K.overlay=null,!0)}function He(){if(K.wrapper&&!oe.isPrintingPDF()){if(!w.disableLayout){g&&!w.embedded&&document.documentElement.style.setProperty("--vh",.01*window.innerHeight+"px");const e=Be(),t=U;qe(w.width,w.height),K.slides.style.width=e.width+"px",K.slides.style.height=e.height+"px",U=Math.min(e.presentationWidth/e.width,e.presentationHeight/e.height),U=Math.max(U,w.minScale),U=Math.min(U,w.maxScale),1===U?(K.slides.style.zoom="",K.slides.style.left="",K.slides.style.top="",K.slides.style.bottom="",K.slides.style.right="",xe({layout:""})):(K.slides.style.zoom="",K.slides.style.left="50%",K.slides.style.top="50%",K.slides.style.bottom="auto",K.slides.style.right="auto",xe({layout:"translate(-50%, -50%) scale("+U+")"}));const i=Array.from(K.wrapper.querySelectorAll(".slides section"));for(let t=0,s=i.length;t .stretch, section > .r-stretch").forEach((t=>{let s=c(t,i);if(/(img|video)/gi.test(t.nodeName)){const i=t.naturalWidth||t.videoWidth,a=t.naturalHeight||t.videoHeight,n=Math.min(e/i,s/a);t.style.width=i*n+"px",t.style.height=a*n+"px"}else t.style.width=e+"px",t.style.height=s+"px"}))}function Be(e,t){const i={width:w.width,height:w.height,presentationWidth:e||K.wrapper.offsetWidth,presentationHeight:t||K.wrapper.offsetHeight};return i.presentationWidth-=i.presentationWidth*w.margin,i.presentationHeight-=i.presentationHeight*w.margin,"string"==typeof i.width&&/%$/.test(i.width)&&(i.width=parseInt(i.width,10)/100*i.presentationWidth),"string"==typeof i.height&&/%$/.test(i.height)&&(i.height=parseInt(i.height,10)/100*i.presentationHeight),i}function Oe(e,t){"object"==typeof e&&"function"==typeof e.setAttribute&&e.setAttribute("data-previous-indexv",t||0)}function Ue(e){if("object"==typeof e&&"function"==typeof e.setAttribute&&e.classList.contains("stack")){const t=e.hasAttribute("data-start-indexv")?"data-start-indexv":"data-previous-indexv";return parseInt(e.getAttribute(t)||0,10)}return 0}function We(e=m){return e&&e.parentNode&&!!e.parentNode.nodeName.match(/section/i)}function Ke(){return!(!m||!We(m))&&!m.nextElementSibling}function Ve(){return 0===u&&0===v}function $e(){return!!m&&(!m.nextElementSibling&&(!We(m)||!m.parentNode.nextElementSibling))}function je(){if(w.pause){const e=K.wrapper.classList.contains("paused");Rt(),K.wrapper.classList.add("paused"),!1===e&&Pe({type:"paused"})}}function Xe(){const e=K.wrapper.classList.contains("paused");K.wrapper.classList.remove("paused"),Et(),e&&Pe({type:"resumed"})}function Ye(e){"boolean"==typeof e?e?je():Xe():_e()?Xe():je()}function _e(){return K.wrapper.classList.contains("paused")}function Je(e){"boolean"==typeof e?e?At():St():Y?At():St()}function Qe(){return!(!$||Y)}function Ze(e,t,i,s){if(Pe({type:"beforeslidechange",data:{indexh:void 0===e?u:e,indexv:void 0===t?v:t,origin:s}}).defaultPrevented)return;p=m;const a=K.wrapper.querySelectorAll(".slides>section");if(0===a.length)return;void 0!==t||ee.isActive()||(t=Ue(a[e])),p&&p.parentNode&&p.parentNode.classList.contains("stack")&&Oe(p.parentNode,v);const n=I.concat();I.length=0;let r=u||0,o=v||0;u=st(".slides>section",void 0===e?u:e),v=st(".slides>section.present>section",void 0===t?v:t);let l=u!==r||v!==o;l||(p=null);let d=a[u],c=d.querySelectorAll("section");m=c[v]||d;let h=!1;l&&p&&m&&!ee.isActive()&&(p.hasAttribute("data-auto-animate")&&m.hasAttribute("data-auto-animate")&&p.getAttribute("data-auto-animate-id")===m.getAttribute("data-auto-animate-id")&&!(u>r||v>o?m:p).hasAttribute("data-auto-animate-restart")&&(h=!0,K.slides.classList.add("disable-slide-transitions")),V="running"),at(),He(),ee.isActive()&&ee.update(),void 0!==i&&G.goto(i),p&&p!==m&&(p.classList.remove("present"),p.setAttribute("aria-hidden","true"),Ve()&&setTimeout((()=>{ut().forEach((e=>{Oe(e,0)}))}),0));e:for(let e=0,t=I.length;e{fe(be(m))})),ae.update(),se.update(),ce.update(),Z.update(),Z.updateParallax(),J.update(),G.update(),ie.writeURL(),Et(),h&&(setTimeout((()=>{K.slides.classList.remove("disable-slide-transitions")}),0),w.autoAnimate&&Q.run(p,m))}function Ge(){Ae(),Se(),He(),$=w.autoSlide,Et(),Z.create(),ie.writeURL(),G.sortAll(),se.update(),ae.update(),at(),ce.update(),ce.updateVisibility(),Z.update(!0),J.update(),_.formatEmbeddedContent(),!1===w.autoPlayMedia?_.stopEmbeddedContent(m,{unloadIframes:!1}):_.startEmbeddedContent(m),ee.isActive()&&ee.layout()}function et(e=m){Z.sync(e),G.sync(e),_.load(e),Z.update(),ce.update()}function tt(){ct().forEach((e=>{t(e,"section").forEach(((e,t)=>{t>0&&(e.classList.remove("present"),e.classList.remove("past"),e.classList.add("future"),e.setAttribute("aria-hidden","true"))}))}))}function it(e=ct()){e.forEach(((t,i)=>{let s=e[Math.floor(Math.random()*e.length)];s.parentNode===t.parentNode&&t.parentNode.insertBefore(t,s);let a=t.querySelectorAll("section");a.length&&it(a)}))}function st(e,i){let s=t(K.wrapper,e),a=s.length,n=oe.isPrintingPDF();if(a){w.loop&&(i%=a)<0&&(i=a+i),i=Math.max(Math.min(i,a-1),0);for(let e=0;e{e.classList.add("visible"),e.classList.remove("current-fragment")}))):e>i&&(a.classList.add(r?"past":"future"),w.fragments&&t(a,".fragment.visible").forEach((e=>{e.classList.remove("visible","current-fragment")})))}let e=s[i],r=e.classList.contains("present");e.classList.add("present"),e.removeAttribute("hidden"),e.removeAttribute("aria-hidden"),r||Pe({target:e,type:"visible",bubbles:!1});let o=e.getAttribute("data-state");o&&(I=I.concat(o.split(" ")))}else i=0;return i}function at(){let e,i,s=ct(),a=s.length;if(a&&void 0!==u){let n=ee.isActive()?10:w.viewDistance;g&&(n=ee.isActive()?6:w.mobileViewDistance),oe.isPrintingPDF()&&(n=Number.MAX_VALUE);for(let r=0;rsection"),i=K.wrapper.querySelectorAll(".slides>section.present>section"),s={left:u>0,right:u0,down:v1&&(s.left=!0,s.right=!0),i.length>1&&(s.up=!0,s.down=!0)),t.length>1&&"linear"===w.navigationMode&&(s.right=s.right||s.down,s.left=s.left||s.up),!0===e){let e=G.availableRoutes();s.left=s.left||e.prev,s.up=s.up||e.prev,s.down=s.down||e.next,s.right=s.right||e.next}if(w.rtl){let e=s.left;s.left=s.right,s.right=e}return s}function rt(e=m){let t=ct(),i=0;e:for(let s=0;s0){let i=.9;t+=m.querySelectorAll(".fragment.visible").length/e.length*i}}return Math.min(t/(e-1),1)}function lt(e){let i,s=u,a=v;if(e){let i=We(e),n=i?e.parentNode:e,r=ct();s=Math.max(r.indexOf(n),0),a=void 0,i&&(a=Math.max(t(e.parentNode,"section").indexOf(e),0))}if(!e&&m){if(m.querySelectorAll(".fragment").length>0){let e=m.querySelector(".current-fragment");i=e&&e.hasAttribute("data-fragment-index")?parseInt(e.getAttribute("data-fragment-index"),10):m.querySelectorAll(".fragment.visible").length-1}}return{h:s,v:a,f:i}}function dt(){return t(K.wrapper,'.slides section:not(.stack):not([data-visibility="uncounted"])')}function ct(){return t(K.wrapper,".slides>section")}function ht(){return t(K.wrapper,".slides>section>section")}function ut(){return t(K.wrapper,".slides>section.stack")}function gt(){return ct().length>1}function vt(){return ht().length>1}function pt(){return dt().map((e=>{let t={};for(let i=0;i{e.hasAttribute("data-autoplay")&&$&&1e3*e.duration/e.playbackRate>$&&($=1e3*e.duration/e.playbackRate+1e3)}))),!$||Y||_e()||ee.isActive()||$e()&&!G.availableRoutes().next&&!0!==w.loop||(j=setTimeout((()=>{"function"==typeof w.autoSlideMethod?w.autoSlideMethod():Nt(),Et()}),$),X=Date.now()),f&&f.setPlaying(-1!==j)}}function Rt(){clearTimeout(j),j=-1}function St(){$&&!Y&&(Y=!0,Pe({type:"autoslidepaused"}),clearTimeout(j),f&&f.setPlaying(!1))}function At(){$&&Y&&(Y=!1,Pe({type:"autoslideresumed"}),Et())}function kt({skipFragments:e=!1}={}){A.hasNavigatedHorizontally=!0,w.rtl?(ee.isActive()||e||!1===G.next())&&nt().left&&Ze(u+1,"grid"===w.navigationMode?v:void 0):(ee.isActive()||e||!1===G.prev())&&nt().left&&Ze(u-1,"grid"===w.navigationMode?v:void 0)}function Lt({skipFragments:e=!1}={}){A.hasNavigatedHorizontally=!0,w.rtl?(ee.isActive()||e||!1===G.prev())&&nt().right&&Ze(u-1,"grid"===w.navigationMode?v:void 0):(ee.isActive()||e||!1===G.next())&&nt().right&&Ze(u+1,"grid"===w.navigationMode?v:void 0)}function Ct({skipFragments:e=!1}={}){(ee.isActive()||e||!1===G.prev())&&nt().up&&Ze(u,v-1)}function xt({skipFragments:e=!1}={}){A.hasNavigatedVertically=!0,(ee.isActive()||e||!1===G.next())&&nt().down&&Ze(u,v+1)}function Pt({skipFragments:e=!1}={}){if(e||!1===G.prev())if(nt().up)Ct({skipFragments:e});else{let i;if(i=w.rtl?t(K.wrapper,".slides>section.future").pop():t(K.wrapper,".slides>section.past").pop(),i&&i.classList.contains("stack")){let e=i.querySelectorAll("section").length-1||void 0;Ze(u-1,e)}else kt({skipFragments:e})}}function Nt({skipFragments:e=!1}={}){if(A.hasNavigatedHorizontally=!0,A.hasNavigatedVertically=!0,e||!1===G.next()){let t=nt();t.down&&t.right&&w.loop&&Ke()&&(t.down=!1),t.down?xt({skipFragments:e}):w.rtl?kt({skipFragments:e}):Lt({skipFragments:e})}}function Mt(e){w.autoSlideStoppable&&St()}function Dt(e){let t=e.data;if("string"==typeof t&&"{"===t.charAt(0)&&"}"===t.charAt(t.length-1)&&(t=JSON.parse(t),t.method&&"function"==typeof h[t.method]))if(!1===R.test(t.method)){const e=h[t.method].apply(h,t.args);Ne("callback",{method:t.method,result:e})}else console.warn('reveal.js: "'+t.method+'" is is blacklisted from the postMessage API')}function It(e){"running"===V&&/section/gi.test(e.target.nodeName)&&(V="idle",Pe({type:"slidetransitionend",data:{indexh:u,indexv:v,previousSlide:p,currentSlide:m}}))}function Tt(e){const t=r(e.target,'a[href^="#"]');if(t){const i=t.getAttribute("href"),s=ie.getIndicesFromHash(i);s&&(h.slide(s.h,s.v,s.f),e.preventDefault())}}function Ft(e){He()}function zt(e){!1===document.hidden&&document.activeElement!==document.body&&("function"==typeof document.activeElement.blur&&document.activeElement.blur(),document.body.focus())}function Ht(e){(document.fullscreenElement||document.webkitFullscreenElement)===K.wrapper&&(e.stopImmediatePropagation(),setTimeout((()=>{h.layout(),h.focus.focus()}),1))}function qt(e){if(e.currentTarget&&e.currentTarget.hasAttribute("href")){let t=e.currentTarget.getAttribute("href");t&&(Ie(t),e.preventDefault())}}function Bt(e){$e()&&!1===w.loop?(Ze(0,0),At()):Y?At():St()}const Ot={VERSION:"4.3.1",initialize:he,configure:Re,destroy:ke,sync:Ge,syncSlide:et,syncFragments:G.sync.bind(G),slide:Ze,left:kt,right:Lt,up:Ct,down:xt,prev:Pt,next:Nt,navigateLeft:kt,navigateRight:Lt,navigateUp:Ct,navigateDown:xt,navigatePrev:Pt,navigateNext:Nt,navigateFragment:G.goto.bind(G),prevFragment:G.prev.bind(G),nextFragment:G.next.bind(G),on:Le,off:Ce,addEventListener:Le,removeEventListener:Ce,layout:He,shuffle:it,availableRoutes:nt,availableFragments:G.availableRoutes.bind(G),toggleHelp:Te,toggleOverview:ee.toggle.bind(ee),togglePause:Ye,toggleAutoSlide:Je,isFirstSlide:Ve,isLastSlide:$e,isLastVerticalSlide:Ke,isVerticalSlide:We,isPaused:_e,isAutoSliding:Qe,isSpeakerNotes:ce.isSpeakerNotesWindow.bind(ce),isOverview:ee.isActive.bind(ee),isFocused:le.isFocused.bind(le),isPrintingPDF:oe.isPrintingPDF.bind(oe),isReady:()=>S,loadSlide:_.load.bind(_),unloadSlide:_.unload.bind(_),showPreview:Ie,hidePreview:ze,addEventListeners:Se,removeEventListeners:Ae,dispatchEvent:Pe,getState:yt,setState:wt,getProgress:ot,getIndices:lt,getSlidesAttributes:pt,getSlidePastCount:rt,getTotalSlides:mt,getSlide:ft,getPreviousSlide:()=>p,getCurrentSlide:()=>m,getSlideBackground:bt,getSlideNotes:ce.getSlideNotes.bind(ce),getSlides:dt,getHorizontalSlides:ct,getVerticalSlides:ht,hasHorizontalSlides:gt,hasVerticalSlides:vt,hasNavigatedHorizontally:()=>A.hasNavigatedHorizontally,hasNavigatedVertically:()=>A.hasNavigatedVertically,addKeyBinding:te.addKeyBinding.bind(te),removeKeyBinding:te.removeKeyBinding.bind(te),triggerKey:te.triggerKey.bind(te),registerKeyboardShortcut:te.registerKeyboardShortcut.bind(te),getComputedSlideSize:Be,getScale:()=>U,getConfig:()=>w,getQueryHash:d,getSlidePath:ie.getHash.bind(ie),getRevealElement:()=>n,getSlidesElement:()=>K.slides,getViewportElement:()=>K.viewport,getBackgroundsElement:()=>Z.element,registerPlugin:re.registerPlugin.bind(re),hasPlugin:re.hasPlugin.bind(re),getPlugin:re.getPlugin.bind(re),getPlugins:re.getRegisteredPlugins.bind(re)};return e(h,{...Ot,announceStatus:fe,getStatusText:be,print:oe,focus:le,progress:ae,controls:se,location:ie,overview:ee,fragments:G,slideContent:_,slideNumber:J,onUserInput:Mt,closeOverlay:ze,updateSlidesVisibility:at,layoutSlideContents:qe,transformSlides:xe,cueAutoSlide:Et,cancelAutoSlide:Rt}),Ot}let W=U,K=[];W.initialize=e=>(Object.assign(W,new U(document.querySelector(".reveal"),e)),K.map((e=>e(W))),W.initialize()),["configure","on","off","addEventListener","removeEventListener","registerPlugin"].forEach((e=>{W[e]=(...t)=>{K.push((i=>i[e].call(null,...t)))}})),W.isReady=()=>!1,W.VERSION="4.3.1";export default W;
+//# sourceMappingURL=reveal.esm.js.map
diff --git a/slides_files/libs/revealjs/dist/reveal.esm.js.map b/slides_files/libs/revealjs/dist/reveal.esm.js.map
new file mode 100644
index 000000000..286c75a21
--- /dev/null
+++ b/slides_files/libs/revealjs/dist/reveal.esm.js.map
@@ -0,0 +1 @@
+{"version":3,"file":"reveal.esm.js","sources":["../js/utils/util.js","../js/utils/device.js","../node_modules/fitty/dist/fitty.module.js","../js/controllers/slidecontent.js","../js/controllers/slidenumber.js","../js/utils/color.js","../js/controllers/backgrounds.js","../js/utils/constants.js","../js/controllers/autoanimate.js","../js/controllers/fragments.js","../js/controllers/overview.js","../js/controllers/keyboard.js","../js/controllers/location.js","../js/controllers/controls.js","../js/controllers/progress.js","../js/controllers/pointer.js","../js/utils/loader.js","../js/controllers/plugins.js","../js/controllers/print.js","../js/controllers/touch.js","../js/controllers/focus.js","../js/controllers/notes.js","../js/components/playback.js","../js/config.js","../js/reveal.js","../js/index.js"],"sourcesContent":["/**\n * Extend object a with the properties of object b.\n * If there's a conflict, object b takes precedence.\n *\n * @param {object} a\n * @param {object} b\n */\nexport const extend = ( a, b ) => {\n\n\tfor( let i in b ) {\n\t\ta[ i ] = b[ i ];\n\t}\n\n\treturn a;\n\n}\n\n/**\n * querySelectorAll but returns an Array.\n */\nexport const queryAll = ( el, selector ) => {\n\n\treturn Array.from( el.querySelectorAll( selector ) );\n\n}\n\n/**\n * classList.toggle() with cross browser support\n */\nexport const toggleClass = ( el, className, value ) => {\n\tif( value ) {\n\t\tel.classList.add( className );\n\t}\n\telse {\n\t\tel.classList.remove( className );\n\t}\n}\n\n/**\n * Utility for deserializing a value.\n *\n * @param {*} value\n * @return {*}\n */\nexport const deserialize = ( value ) => {\n\n\tif( typeof value === 'string' ) {\n\t\tif( value === 'null' ) return null;\n\t\telse if( value === 'true' ) return true;\n\t\telse if( value === 'false' ) return false;\n\t\telse if( value.match( /^-?[\\d\\.]+$/ ) ) return parseFloat( value );\n\t}\n\n\treturn value;\n\n}\n\n/**\n * Measures the distance in pixels between point a\n * and point b.\n *\n * @param {object} a point with x/y properties\n * @param {object} b point with x/y properties\n *\n * @return {number}\n */\nexport const distanceBetween = ( a, b ) => {\n\n\tlet dx = a.x - b.x,\n\t\tdy = a.y - b.y;\n\n\treturn Math.sqrt( dx*dx + dy*dy );\n\n}\n\n/**\n * Applies a CSS transform to the target element.\n *\n * @param {HTMLElement} element\n * @param {string} transform\n */\nexport const transformElement = ( element, transform ) => {\n\n\telement.style.transform = transform;\n\n}\n\n/**\n * Element.matches with IE support.\n *\n * @param {HTMLElement} target The element to match\n * @param {String} selector The CSS selector to match\n * the element against\n *\n * @return {Boolean}\n */\nexport const matches = ( target, selector ) => {\n\n\tlet matchesMethod = target.matches || target.matchesSelector || target.msMatchesSelector;\n\n\treturn !!( matchesMethod && matchesMethod.call( target, selector ) );\n\n}\n\n/**\n * Find the closest parent that matches the given\n * selector.\n *\n * @param {HTMLElement} target The child element\n * @param {String} selector The CSS selector to match\n * the parents against\n *\n * @return {HTMLElement} The matched parent or null\n * if no matching parent was found\n */\nexport const closest = ( target, selector ) => {\n\n\t// Native Element.closest\n\tif( typeof target.closest === 'function' ) {\n\t\treturn target.closest( selector );\n\t}\n\n\t// Polyfill\n\twhile( target ) {\n\t\tif( matches( target, selector ) ) {\n\t\t\treturn target;\n\t\t}\n\n\t\t// Keep searching\n\t\ttarget = target.parentNode;\n\t}\n\n\treturn null;\n\n}\n\n/**\n * Handling the fullscreen functionality via the fullscreen API\n *\n * @see http://fullscreen.spec.whatwg.org/\n * @see https://developer.mozilla.org/en-US/docs/DOM/Using_fullscreen_mode\n */\nexport const enterFullscreen = element => {\n\n\telement = element || document.documentElement;\n\n\t// Check which implementation is available\n\tlet requestMethod = element.requestFullscreen ||\n\t\t\t\t\t\telement.webkitRequestFullscreen ||\n\t\t\t\t\t\telement.webkitRequestFullScreen ||\n\t\t\t\t\t\telement.mozRequestFullScreen ||\n\t\t\t\t\t\telement.msRequestFullscreen;\n\n\tif( requestMethod ) {\n\t\trequestMethod.apply( element );\n\t}\n\n}\n\n/**\n * Creates an HTML element and returns a reference to it.\n * If the element already exists the existing instance will\n * be returned.\n *\n * @param {HTMLElement} container\n * @param {string} tagname\n * @param {string} classname\n * @param {string} innerHTML\n *\n * @return {HTMLElement}\n */\nexport const createSingletonNode = ( container, tagname, classname, innerHTML='' ) => {\n\n\t// Find all nodes matching the description\n\tlet nodes = container.querySelectorAll( '.' + classname );\n\n\t// Check all matches to find one which is a direct child of\n\t// the specified container\n\tfor( let i = 0; i < nodes.length; i++ ) {\n\t\tlet testNode = nodes[i];\n\t\tif( testNode.parentNode === container ) {\n\t\t\treturn testNode;\n\t\t}\n\t}\n\n\t// If no node was found, create it now\n\tlet node = document.createElement( tagname );\n\tnode.className = classname;\n\tnode.innerHTML = innerHTML;\n\tcontainer.appendChild( node );\n\n\treturn node;\n\n}\n\n/**\n * Injects the given CSS styles into the DOM.\n *\n * @param {string} value\n */\nexport const createStyleSheet = ( value ) => {\n\n\tlet tag = document.createElement( 'style' );\n\ttag.type = 'text/css';\n\n\tif( value && value.length > 0 ) {\n\t\tif( tag.styleSheet ) {\n\t\t\ttag.styleSheet.cssText = value;\n\t\t}\n\t\telse {\n\t\t\ttag.appendChild( document.createTextNode( value ) );\n\t\t}\n\t}\n\n\tdocument.head.appendChild( tag );\n\n\treturn tag;\n\n}\n\n/**\n * Returns a key:value hash of all query params.\n */\nexport const getQueryHash = () => {\n\n\tlet query = {};\n\n\tlocation.search.replace( /[A-Z0-9]+?=([\\w\\.%-]*)/gi, a => {\n\t\tquery[ a.split( '=' ).shift() ] = a.split( '=' ).pop();\n\t} );\n\n\t// Basic deserialization\n\tfor( let i in query ) {\n\t\tlet value = query[ i ];\n\n\t\tquery[ i ] = deserialize( unescape( value ) );\n\t}\n\n\t// Do not accept new dependencies via query config to avoid\n\t// the potential of malicious script injection\n\tif( typeof query['dependencies'] !== 'undefined' ) delete query['dependencies'];\n\n\treturn query;\n\n}\n\n/**\n * Returns the remaining height within the parent of the\n * target element.\n *\n * remaining height = [ configured parent height ] - [ current parent height ]\n *\n * @param {HTMLElement} element\n * @param {number} [height]\n */\nexport const getRemainingHeight = ( element, height = 0 ) => {\n\n\tif( element ) {\n\t\tlet newHeight, oldHeight = element.style.height;\n\n\t\t// Change the .stretch element height to 0 in order find the height of all\n\t\t// the other elements\n\t\telement.style.height = '0px';\n\n\t\t// In Overview mode, the parent (.slide) height is set of 700px.\n\t\t// Restore it temporarily to its natural height.\n\t\telement.parentNode.style.height = 'auto';\n\n\t\tnewHeight = height - element.parentNode.offsetHeight;\n\n\t\t// Restore the old height, just in case\n\t\telement.style.height = oldHeight + 'px';\n\n\t\t// Clear the parent (.slide) height. .removeProperty works in IE9+\n\t\telement.parentNode.style.removeProperty('height');\n\n\t\treturn newHeight;\n\t}\n\n\treturn height;\n\n}\n\nconst fileExtensionToMimeMap = {\n\t'mp4': 'video/mp4',\n\t'm4a': 'video/mp4',\n\t'ogv': 'video/ogg',\n\t'mpeg': 'video/mpeg',\n\t'webm': 'video/webm'\n}\n\n/**\n * Guess the MIME type for common file formats.\n */\nexport const getMimeTypeFromFile = ( filename='' ) => {\n\treturn fileExtensionToMimeMap[filename.split('.').pop()]\n}","const UA = navigator.userAgent;\n\nexport const isMobile = /(iphone|ipod|ipad|android)/gi.test( UA ) ||\n\t\t\t\t\t\t( navigator.platform === 'MacIntel' && navigator.maxTouchPoints > 1 ); // iPadOS\n\nexport const isChrome = /chrome/i.test( UA ) && !/edge/i.test( UA );\n\nexport const isAndroid = /android/gi.test( UA );","/*\n * fitty v2.3.3 - Snugly resizes text to fit its parent container\n * Copyright (c) 2020 Rik Schennink (https://pqina.nl/)\n */\n'use strict';\n\nObject.defineProperty(exports, \"__esModule\", {\n value: true\n});\n\nvar _extends = Object.assign || function (target) { for (var i = 1; i < arguments.length; i++) { var source = arguments[i]; for (var key in source) { if (Object.prototype.hasOwnProperty.call(source, key)) { target[key] = source[key]; } } } return target; };\n\nexports.default = function (w) {\n\n // no window, early exit\n if (!w) return;\n\n // node list to array helper method\n var toArray = function toArray(nl) {\n return [].slice.call(nl);\n };\n\n // states\n var DrawState = {\n IDLE: 0,\n DIRTY_CONTENT: 1,\n DIRTY_LAYOUT: 2,\n DIRTY: 3\n };\n\n // all active fitty elements\n var fitties = [];\n\n // group all redraw calls till next frame, we cancel each frame request when a new one comes in. If no support for request animation frame, this is an empty function and supports for fitty stops.\n var redrawFrame = null;\n var requestRedraw = 'requestAnimationFrame' in w ? function () {\n w.cancelAnimationFrame(redrawFrame);\n redrawFrame = w.requestAnimationFrame(function () {\n return redraw(fitties.filter(function (f) {\n return f.dirty && f.active;\n }));\n });\n } : function () {};\n\n // sets all fitties to dirty so they are redrawn on the next redraw loop, then calls redraw\n var redrawAll = function redrawAll(type) {\n return function () {\n fitties.forEach(function (f) {\n return f.dirty = type;\n });\n requestRedraw();\n };\n };\n\n // redraws fitties so they nicely fit their parent container\n var redraw = function redraw(fitties) {\n\n // getting info from the DOM at this point should not trigger a reflow, let's gather as much intel as possible before triggering a reflow\n\n // check if styles of all fitties have been computed\n fitties.filter(function (f) {\n return !f.styleComputed;\n }).forEach(function (f) {\n f.styleComputed = computeStyle(f);\n });\n\n // restyle elements that require pre-styling, this triggers a reflow, please try to prevent by adding CSS rules (see docs)\n fitties.filter(shouldPreStyle).forEach(applyStyle);\n\n // we now determine which fitties should be redrawn\n var fittiesToRedraw = fitties.filter(shouldRedraw);\n\n // we calculate final styles for these fitties\n fittiesToRedraw.forEach(calculateStyles);\n\n // now we apply the calculated styles from our previous loop\n fittiesToRedraw.forEach(function (f) {\n applyStyle(f);\n markAsClean(f);\n });\n\n // now we dispatch events for all restyled fitties\n fittiesToRedraw.forEach(dispatchFitEvent);\n };\n\n var markAsClean = function markAsClean(f) {\n return f.dirty = DrawState.IDLE;\n };\n\n var calculateStyles = function calculateStyles(f) {\n\n // get available width from parent node\n f.availableWidth = f.element.parentNode.clientWidth;\n\n // the space our target element uses\n f.currentWidth = f.element.scrollWidth;\n\n // remember current font size\n f.previousFontSize = f.currentFontSize;\n\n // let's calculate the new font size\n f.currentFontSize = Math.min(Math.max(f.minSize, f.availableWidth / f.currentWidth * f.previousFontSize), f.maxSize);\n\n // if allows wrapping, only wrap when at minimum font size (otherwise would break container)\n f.whiteSpace = f.multiLine && f.currentFontSize === f.minSize ? 'normal' : 'nowrap';\n };\n\n // should always redraw if is not dirty layout, if is dirty layout, only redraw if size has changed\n var shouldRedraw = function shouldRedraw(f) {\n return f.dirty !== DrawState.DIRTY_LAYOUT || f.dirty === DrawState.DIRTY_LAYOUT && f.element.parentNode.clientWidth !== f.availableWidth;\n };\n\n // every fitty element is tested for invalid styles\n var computeStyle = function computeStyle(f) {\n\n // get style properties\n var style = w.getComputedStyle(f.element, null);\n\n // get current font size in pixels (if we already calculated it, use the calculated version)\n f.currentFontSize = parseFloat(style.getPropertyValue('font-size'));\n\n // get display type and wrap mode\n f.display = style.getPropertyValue('display');\n f.whiteSpace = style.getPropertyValue('white-space');\n };\n\n // determines if this fitty requires initial styling, can be prevented by applying correct styles through CSS\n var shouldPreStyle = function shouldPreStyle(f) {\n\n var preStyle = false;\n\n // if we already tested for prestyling we don't have to do it again\n if (f.preStyleTestCompleted) return false;\n\n // should have an inline style, if not, apply\n if (!/inline-/.test(f.display)) {\n preStyle = true;\n f.display = 'inline-block';\n }\n\n // to correctly calculate dimensions the element should have whiteSpace set to nowrap\n if (f.whiteSpace !== 'nowrap') {\n preStyle = true;\n f.whiteSpace = 'nowrap';\n }\n\n // we don't have to do this twice\n f.preStyleTestCompleted = true;\n\n return preStyle;\n };\n\n // apply styles to single fitty\n var applyStyle = function applyStyle(f) {\n f.element.style.whiteSpace = f.whiteSpace;\n f.element.style.display = f.display;\n f.element.style.fontSize = f.currentFontSize + 'px';\n };\n\n // dispatch a fit event on a fitty\n var dispatchFitEvent = function dispatchFitEvent(f) {\n f.element.dispatchEvent(new CustomEvent('fit', {\n detail: {\n oldValue: f.previousFontSize,\n newValue: f.currentFontSize,\n scaleFactor: f.currentFontSize / f.previousFontSize\n }\n }));\n };\n\n // fit method, marks the fitty as dirty and requests a redraw (this will also redraw any other fitty marked as dirty)\n var fit = function fit(f, type) {\n return function () {\n f.dirty = type;\n if (!f.active) return;\n requestRedraw();\n };\n };\n\n var init = function init(f) {\n\n // save some of the original CSS properties before we change them\n f.originalStyle = {\n whiteSpace: f.element.style.whiteSpace,\n display: f.element.style.display,\n fontSize: f.element.style.fontSize\n };\n\n // should we observe DOM mutations\n observeMutations(f);\n\n // this is a new fitty so we need to validate if it's styles are in order\n f.newbie = true;\n\n // because it's a new fitty it should also be dirty, we want it to redraw on the first loop\n f.dirty = true;\n\n // we want to be able to update this fitty\n fitties.push(f);\n };\n\n var destroy = function destroy(f) {\n return function () {\n\n // remove from fitties array\n fitties = fitties.filter(function (_) {\n return _.element !== f.element;\n });\n\n // stop observing DOM\n if (f.observeMutations) f.observer.disconnect();\n\n // reset the CSS properties we changes\n f.element.style.whiteSpace = f.originalStyle.whiteSpace;\n f.element.style.display = f.originalStyle.display;\n f.element.style.fontSize = f.originalStyle.fontSize;\n };\n };\n\n // add a new fitty, does not redraw said fitty\n var subscribe = function subscribe(f) {\n return function () {\n if (f.active) return;\n f.active = true;\n requestRedraw();\n };\n };\n\n // remove an existing fitty\n var unsubscribe = function unsubscribe(f) {\n return function () {\n return f.active = false;\n };\n };\n\n var observeMutations = function observeMutations(f) {\n\n // no observing?\n if (!f.observeMutations) return;\n\n // start observing mutations\n f.observer = new MutationObserver(fit(f, DrawState.DIRTY_CONTENT));\n\n // start observing\n f.observer.observe(f.element, f.observeMutations);\n };\n\n // default mutation observer settings\n var mutationObserverDefaultSetting = {\n subtree: true,\n childList: true,\n characterData: true\n };\n\n // default fitty options\n var defaultOptions = {\n minSize: 16,\n maxSize: 512,\n multiLine: true,\n observeMutations: 'MutationObserver' in w ? mutationObserverDefaultSetting : false\n };\n\n // array of elements in, fitty instances out\n function fittyCreate(elements, options) {\n\n // set options object\n var fittyOptions = _extends({}, defaultOptions, options);\n\n // create fitties\n var publicFitties = elements.map(function (element) {\n\n // create fitty instance\n var f = _extends({}, fittyOptions, {\n\n // internal options for this fitty\n element: element,\n active: true\n });\n\n // initialise this fitty\n init(f);\n\n // expose API\n return {\n element: element,\n fit: fit(f, DrawState.DIRTY),\n unfreeze: subscribe(f),\n freeze: unsubscribe(f),\n unsubscribe: destroy(f)\n };\n });\n\n // call redraw on newly initiated fitties\n requestRedraw();\n\n // expose fitties\n return publicFitties;\n }\n\n // fitty creation function\n function fitty(target) {\n var options = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : {};\n\n\n // if target is a string\n return typeof target === 'string' ?\n\n // treat it as a querySelector\n fittyCreate(toArray(document.querySelectorAll(target)), options) :\n\n // create single fitty\n fittyCreate([target], options)[0];\n }\n\n // handles viewport changes, redraws all fitties, but only does so after a timeout\n var resizeDebounce = null;\n var onWindowResized = function onWindowResized() {\n w.clearTimeout(resizeDebounce);\n resizeDebounce = w.setTimeout(redrawAll(DrawState.DIRTY_LAYOUT), fitty.observeWindowDelay);\n };\n\n // define observe window property, so when we set it to true or false events are automatically added and removed\n var events = ['resize', 'orientationchange'];\n Object.defineProperty(fitty, 'observeWindow', {\n set: function set(enabled) {\n var method = (enabled ? 'add' : 'remove') + 'EventListener';\n events.forEach(function (e) {\n w[method](e, onWindowResized);\n });\n }\n });\n\n // fitty global properties (by setting observeWindow to true the events above get added)\n fitty.observeWindow = true;\n fitty.observeWindowDelay = 100;\n\n // public fit all method, will force redraw no matter what\n fitty.fitAll = redrawAll(DrawState.DIRTY);\n\n // export our fitty function, we don't want to keep it to our selves\n return fitty;\n}(typeof window === 'undefined' ? null : window);","import { extend, queryAll, closest, getMimeTypeFromFile } from '../utils/util.js'\nimport { isMobile } from '../utils/device.js'\n\nimport fitty from 'fitty';\n\n/**\n * Handles loading, unloading and playback of slide\n * content such as images, videos and iframes.\n */\nexport default class SlideContent {\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t\tthis.startEmbeddedIframe = this.startEmbeddedIframe.bind( this );\n\n\t}\n\n\t/**\n\t * Should the given element be preloaded?\n\t * Decides based on local element attributes and global config.\n\t *\n\t * @param {HTMLElement} element\n\t */\n\tshouldPreload( element ) {\n\n\t\t// Prefer an explicit global preload setting\n\t\tlet preload = this.Reveal.getConfig().preloadIframes;\n\n\t\t// If no global setting is available, fall back on the element's\n\t\t// own preload setting\n\t\tif( typeof preload !== 'boolean' ) {\n\t\t\tpreload = element.hasAttribute( 'data-preload' );\n\t\t}\n\n\t\treturn preload;\n\t}\n\n\t/**\n\t * Called when the given slide is within the configured view\n\t * distance. Shows the slide element and loads any content\n\t * that is set to load lazily (data-src).\n\t *\n\t * @param {HTMLElement} slide Slide to show\n\t */\n\tload( slide, options = {} ) {\n\n\t\t// Show the slide element\n\t\tslide.style.display = this.Reveal.getConfig().display;\n\n\t\t// Media elements with data-src attributes\n\t\tqueryAll( slide, 'img[data-src], video[data-src], audio[data-src], iframe[data-src]' ).forEach( element => {\n\t\t\tif( element.tagName !== 'IFRAME' || this.shouldPreload( element ) ) {\n\t\t\t\telement.setAttribute( 'src', element.getAttribute( 'data-src' ) );\n\t\t\t\telement.setAttribute( 'data-lazy-loaded', '' );\n\t\t\t\telement.removeAttribute( 'data-src' );\n\t\t\t}\n\t\t} );\n\n\t\t// Media elements with children\n\t\tqueryAll( slide, 'video, audio' ).forEach( media => {\n\t\t\tlet sources = 0;\n\n\t\t\tqueryAll( media, 'source[data-src]' ).forEach( source => {\n\t\t\t\tsource.setAttribute( 'src', source.getAttribute( 'data-src' ) );\n\t\t\t\tsource.removeAttribute( 'data-src' );\n\t\t\t\tsource.setAttribute( 'data-lazy-loaded', '' );\n\t\t\t\tsources += 1;\n\t\t\t} );\n\n\t\t\t// Enable inline video playback in mobile Safari\n\t\t\tif( isMobile && media.tagName === 'VIDEO' ) {\n\t\t\t\tmedia.setAttribute( 'playsinline', '' );\n\t\t\t}\n\n\t\t\t// If we rewrote sources for this video/audio element, we need\n\t\t\t// to manually tell it to load from its new origin\n\t\t\tif( sources > 0 ) {\n\t\t\t\tmedia.load();\n\t\t\t}\n\t\t} );\n\n\n\t\t// Show the corresponding background element\n\t\tlet background = slide.slideBackgroundElement;\n\t\tif( background ) {\n\t\t\tbackground.style.display = 'block';\n\n\t\t\tlet backgroundContent = slide.slideBackgroundContentElement;\n\t\t\tlet backgroundIframe = slide.getAttribute( 'data-background-iframe' );\n\n\t\t\t// If the background contains media, load it\n\t\t\tif( background.hasAttribute( 'data-loaded' ) === false ) {\n\t\t\t\tbackground.setAttribute( 'data-loaded', 'true' );\n\n\t\t\t\tlet backgroundImage = slide.getAttribute( 'data-background-image' ),\n\t\t\t\t\tbackgroundVideo = slide.getAttribute( 'data-background-video' ),\n\t\t\t\t\tbackgroundVideoLoop = slide.hasAttribute( 'data-background-video-loop' ),\n\t\t\t\t\tbackgroundVideoMuted = slide.hasAttribute( 'data-background-video-muted' );\n\n\t\t\t\t// Images\n\t\t\t\tif( backgroundImage ) {\n\t\t\t\t\t// base64\n\t\t\t\t\tif( /^data:/.test( backgroundImage.trim() ) ) {\n\t\t\t\t\t\tbackgroundContent.style.backgroundImage = `url(${backgroundImage.trim()})`;\n\t\t\t\t\t}\n\t\t\t\t\t// URL(s)\n\t\t\t\t\telse {\n\t\t\t\t\t\tbackgroundContent.style.backgroundImage = backgroundImage.split( ',' ).map( background => {\n\t\t\t\t\t\t\treturn `url(${encodeURI(background.trim())})`;\n\t\t\t\t\t\t}).join( ',' );\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\t// Videos\n\t\t\t\telse if ( backgroundVideo && !this.Reveal.isSpeakerNotes() ) {\n\t\t\t\t\tlet video = document.createElement( 'video' );\n\n\t\t\t\t\tif( backgroundVideoLoop ) {\n\t\t\t\t\t\tvideo.setAttribute( 'loop', '' );\n\t\t\t\t\t}\n\n\t\t\t\t\tif( backgroundVideoMuted ) {\n\t\t\t\t\t\tvideo.muted = true;\n\t\t\t\t\t}\n\n\t\t\t\t\t// Enable inline playback in mobile Safari\n\t\t\t\t\t//\n\t\t\t\t\t// Mute is required for video to play when using\n\t\t\t\t\t// swipe gestures to navigate since they don't\n\t\t\t\t\t// count as direct user actions :'(\n\t\t\t\t\tif( isMobile ) {\n\t\t\t\t\t\tvideo.muted = true;\n\t\t\t\t\t\tvideo.setAttribute( 'playsinline', '' );\n\t\t\t\t\t}\n\n\t\t\t\t\t// Support comma separated lists of video sources\n\t\t\t\t\tbackgroundVideo.split( ',' ).forEach( source => {\n\t\t\t\t\t\tlet type = getMimeTypeFromFile( source );\n\t\t\t\t\t\tif( type ) {\n\t\t\t\t\t\t\tvideo.innerHTML += ``;\n\t\t\t\t\t\t}\n\t\t\t\t\t\telse {\n\t\t\t\t\t\t\tvideo.innerHTML += ``;\n\t\t\t\t\t\t}\n\t\t\t\t\t} );\n\n\t\t\t\t\tbackgroundContent.appendChild( video );\n\t\t\t\t}\n\t\t\t\t// Iframes\n\t\t\t\telse if( backgroundIframe && options.excludeIframes !== true ) {\n\t\t\t\t\tlet iframe = document.createElement( 'iframe' );\n\t\t\t\t\tiframe.setAttribute( 'allowfullscreen', '' );\n\t\t\t\t\tiframe.setAttribute( 'mozallowfullscreen', '' );\n\t\t\t\t\tiframe.setAttribute( 'webkitallowfullscreen', '' );\n\t\t\t\t\tiframe.setAttribute( 'allow', 'autoplay' );\n\n\t\t\t\t\tiframe.setAttribute( 'data-src', backgroundIframe );\n\n\t\t\t\t\tiframe.style.width = '100%';\n\t\t\t\t\tiframe.style.height = '100%';\n\t\t\t\t\tiframe.style.maxHeight = '100%';\n\t\t\t\t\tiframe.style.maxWidth = '100%';\n\n\t\t\t\t\tbackgroundContent.appendChild( iframe );\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Start loading preloadable iframes\n\t\t\tlet backgroundIframeElement = backgroundContent.querySelector( 'iframe[data-src]' );\n\t\t\tif( backgroundIframeElement ) {\n\n\t\t\t\t// Check if this iframe is eligible to be preloaded\n\t\t\t\tif( this.shouldPreload( background ) && !/autoplay=(1|true|yes)/gi.test( backgroundIframe ) ) {\n\t\t\t\t\tif( backgroundIframeElement.getAttribute( 'src' ) !== backgroundIframe ) {\n\t\t\t\t\t\tbackgroundIframeElement.setAttribute( 'src', backgroundIframe );\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t}\n\n\t\t}\n\n\t\tthis.layout( slide );\n\n\t}\n\n\t/**\n\t * Applies JS-dependent layout helpers for the given slide,\n\t * if there are any.\n\t */\n\tlayout( slide ) {\n\n\t\t// Autosize text with the r-fit-text class based on the\n\t\t// size of its container. This needs to happen after the\n\t\t// slide is visible in order to measure the text.\n\t\tArray.from( slide.querySelectorAll( '.r-fit-text' ) ).forEach( element => {\n\t\t\tfitty( element, {\n\t\t\t\tminSize: 24,\n\t\t\t\tmaxSize: this.Reveal.getConfig().height * 0.8,\n\t\t\t\tobserveMutations: false,\n\t\t\t\tobserveWindow: false\n\t\t\t} );\n\t\t} );\n\n\t}\n\n\t/**\n\t * Unloads and hides the given slide. This is called when the\n\t * slide is moved outside of the configured view distance.\n\t *\n\t * @param {HTMLElement} slide\n\t */\n\tunload( slide ) {\n\n\t\t// Hide the slide element\n\t\tslide.style.display = 'none';\n\n\t\t// Hide the corresponding background element\n\t\tlet background = this.Reveal.getSlideBackground( slide );\n\t\tif( background ) {\n\t\t\tbackground.style.display = 'none';\n\n\t\t\t// Unload any background iframes\n\t\t\tqueryAll( background, 'iframe[src]' ).forEach( element => {\n\t\t\t\telement.removeAttribute( 'src' );\n\t\t\t} );\n\t\t}\n\n\t\t// Reset lazy-loaded media elements with src attributes\n\t\tqueryAll( slide, 'video[data-lazy-loaded][src], audio[data-lazy-loaded][src], iframe[data-lazy-loaded][src]' ).forEach( element => {\n\t\t\telement.setAttribute( 'data-src', element.getAttribute( 'src' ) );\n\t\t\telement.removeAttribute( 'src' );\n\t\t} );\n\n\t\t// Reset lazy-loaded media elements with children\n\t\tqueryAll( slide, 'video[data-lazy-loaded] source[src], audio source[src]' ).forEach( source => {\n\t\t\tsource.setAttribute( 'data-src', source.getAttribute( 'src' ) );\n\t\t\tsource.removeAttribute( 'src' );\n\t\t} );\n\n\t}\n\n\t/**\n\t * Enforces origin-specific format rules for embedded media.\n\t */\n\tformatEmbeddedContent() {\n\n\t\tlet _appendParamToIframeSource = ( sourceAttribute, sourceURL, param ) => {\n\t\t\tqueryAll( this.Reveal.getSlidesElement(), 'iframe['+ sourceAttribute +'*=\"'+ sourceURL +'\"]' ).forEach( el => {\n\t\t\t\tlet src = el.getAttribute( sourceAttribute );\n\t\t\t\tif( src && src.indexOf( param ) === -1 ) {\n\t\t\t\t\tel.setAttribute( sourceAttribute, src + ( !/\\?/.test( src ) ? '?' : '&' ) + param );\n\t\t\t\t}\n\t\t\t});\n\t\t};\n\n\t\t// YouTube frames must include \"?enablejsapi=1\"\n\t\t_appendParamToIframeSource( 'src', 'youtube.com/embed/', 'enablejsapi=1' );\n\t\t_appendParamToIframeSource( 'data-src', 'youtube.com/embed/', 'enablejsapi=1' );\n\n\t\t// Vimeo frames must include \"?api=1\"\n\t\t_appendParamToIframeSource( 'src', 'player.vimeo.com/', 'api=1' );\n\t\t_appendParamToIframeSource( 'data-src', 'player.vimeo.com/', 'api=1' );\n\n\t}\n\n\t/**\n\t * Start playback of any embedded content inside of\n\t * the given element.\n\t *\n\t * @param {HTMLElement} element\n\t */\n\tstartEmbeddedContent( element ) {\n\n\t\tif( element && !this.Reveal.isSpeakerNotes() ) {\n\n\t\t\t// Restart GIFs\n\t\t\tqueryAll( element, 'img[src$=\".gif\"]' ).forEach( el => {\n\t\t\t\t// Setting the same unchanged source like this was confirmed\n\t\t\t\t// to work in Chrome, FF & Safari\n\t\t\t\tel.setAttribute( 'src', el.getAttribute( 'src' ) );\n\t\t\t} );\n\n\t\t\t// HTML5 media elements\n\t\t\tqueryAll( element, 'video, audio' ).forEach( el => {\n\t\t\t\tif( closest( el, '.fragment' ) && !closest( el, '.fragment.visible' ) ) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\n\t\t\t\t// Prefer an explicit global autoplay setting\n\t\t\t\tlet autoplay = this.Reveal.getConfig().autoPlayMedia;\n\n\t\t\t\t// If no global setting is available, fall back on the element's\n\t\t\t\t// own autoplay setting\n\t\t\t\tif( typeof autoplay !== 'boolean' ) {\n\t\t\t\t\tautoplay = el.hasAttribute( 'data-autoplay' ) || !!closest( el, '.slide-background' );\n\t\t\t\t}\n\n\t\t\t\tif( autoplay && typeof el.play === 'function' ) {\n\n\t\t\t\t\t// If the media is ready, start playback\n\t\t\t\t\tif( el.readyState > 1 ) {\n\t\t\t\t\t\tthis.startEmbeddedMedia( { target: el } );\n\t\t\t\t\t}\n\t\t\t\t\t// Mobile devices never fire a loaded event so instead\n\t\t\t\t\t// of waiting, we initiate playback\n\t\t\t\t\telse if( isMobile ) {\n\t\t\t\t\t\tlet promise = el.play();\n\n\t\t\t\t\t\t// If autoplay does not work, ensure that the controls are visible so\n\t\t\t\t\t\t// that the viewer can start the media on their own\n\t\t\t\t\t\tif( promise && typeof promise.catch === 'function' && el.controls === false ) {\n\t\t\t\t\t\t\tpromise.catch( () => {\n\t\t\t\t\t\t\t\tel.controls = true;\n\n\t\t\t\t\t\t\t\t// Once the video does start playing, hide the controls again\n\t\t\t\t\t\t\t\tel.addEventListener( 'play', () => {\n\t\t\t\t\t\t\t\t\tel.controls = false;\n\t\t\t\t\t\t\t\t} );\n\t\t\t\t\t\t\t} );\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\t// If the media isn't loaded, wait before playing\n\t\t\t\t\telse {\n\t\t\t\t\t\tel.removeEventListener( 'loadeddata', this.startEmbeddedMedia ); // remove first to avoid dupes\n\t\t\t\t\t\tel.addEventListener( 'loadeddata', this.startEmbeddedMedia );\n\t\t\t\t\t}\n\n\t\t\t\t}\n\t\t\t} );\n\n\t\t\t// Normal iframes\n\t\t\tqueryAll( element, 'iframe[src]' ).forEach( el => {\n\t\t\t\tif( closest( el, '.fragment' ) && !closest( el, '.fragment.visible' ) ) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\n\t\t\t\tthis.startEmbeddedIframe( { target: el } );\n\t\t\t} );\n\n\t\t\t// Lazy loading iframes\n\t\t\tqueryAll( element, 'iframe[data-src]' ).forEach( el => {\n\t\t\t\tif( closest( el, '.fragment' ) && !closest( el, '.fragment.visible' ) ) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\n\t\t\t\tif( el.getAttribute( 'src' ) !== el.getAttribute( 'data-src' ) ) {\n\t\t\t\t\tel.removeEventListener( 'load', this.startEmbeddedIframe ); // remove first to avoid dupes\n\t\t\t\t\tel.addEventListener( 'load', this.startEmbeddedIframe );\n\t\t\t\t\tel.setAttribute( 'src', el.getAttribute( 'data-src' ) );\n\t\t\t\t}\n\t\t\t} );\n\n\t\t}\n\n\t}\n\n\t/**\n\t * Starts playing an embedded video/audio element after\n\t * it has finished loading.\n\t *\n\t * @param {object} event\n\t */\n\tstartEmbeddedMedia( event ) {\n\n\t\tlet isAttachedToDOM = !!closest( event.target, 'html' ),\n\t\t\tisVisible \t\t= !!closest( event.target, '.present' );\n\n\t\tif( isAttachedToDOM && isVisible ) {\n\t\t\tevent.target.currentTime = 0;\n\t\t\tevent.target.play();\n\t\t}\n\n\t\tevent.target.removeEventListener( 'loadeddata', this.startEmbeddedMedia );\n\n\t}\n\n\t/**\n\t * \"Starts\" the content of an embedded iframe using the\n\t * postMessage API.\n\t *\n\t * @param {object} event\n\t */\n\tstartEmbeddedIframe( event ) {\n\n\t\tlet iframe = event.target;\n\n\t\tif( iframe && iframe.contentWindow ) {\n\n\t\t\tlet isAttachedToDOM = !!closest( event.target, 'html' ),\n\t\t\t\tisVisible \t\t= !!closest( event.target, '.present' );\n\n\t\t\tif( isAttachedToDOM && isVisible ) {\n\n\t\t\t\t// Prefer an explicit global autoplay setting\n\t\t\t\tlet autoplay = this.Reveal.getConfig().autoPlayMedia;\n\n\t\t\t\t// If no global setting is available, fall back on the element's\n\t\t\t\t// own autoplay setting\n\t\t\t\tif( typeof autoplay !== 'boolean' ) {\n\t\t\t\t\tautoplay = iframe.hasAttribute( 'data-autoplay' ) || !!closest( iframe, '.slide-background' );\n\t\t\t\t}\n\n\t\t\t\t// YouTube postMessage API\n\t\t\t\tif( /youtube\\.com\\/embed\\//.test( iframe.getAttribute( 'src' ) ) && autoplay ) {\n\t\t\t\t\tiframe.contentWindow.postMessage( '{\"event\":\"command\",\"func\":\"playVideo\",\"args\":\"\"}', '*' );\n\t\t\t\t}\n\t\t\t\t// Vimeo postMessage API\n\t\t\t\telse if( /player\\.vimeo\\.com\\//.test( iframe.getAttribute( 'src' ) ) && autoplay ) {\n\t\t\t\t\tiframe.contentWindow.postMessage( '{\"method\":\"play\"}', '*' );\n\t\t\t\t}\n\t\t\t\t// Generic postMessage API\n\t\t\t\telse {\n\t\t\t\t\tiframe.contentWindow.postMessage( 'slide:start', '*' );\n\t\t\t\t}\n\n\t\t\t}\n\n\t\t}\n\n\t}\n\n\t/**\n\t * Stop playback of any embedded content inside of\n\t * the targeted slide.\n\t *\n\t * @param {HTMLElement} element\n\t */\n\tstopEmbeddedContent( element, options = {} ) {\n\n\t\toptions = extend( {\n\t\t\t// Defaults\n\t\t\tunloadIframes: true\n\t\t}, options );\n\n\t\tif( element && element.parentNode ) {\n\t\t\t// HTML5 media elements\n\t\t\tqueryAll( element, 'video, audio' ).forEach( el => {\n\t\t\t\tif( !el.hasAttribute( 'data-ignore' ) && typeof el.pause === 'function' ) {\n\t\t\t\t\tel.setAttribute('data-paused-by-reveal', '');\n\t\t\t\t\tel.pause();\n\t\t\t\t}\n\t\t\t} );\n\n\t\t\t// Generic postMessage API for non-lazy loaded iframes\n\t\t\tqueryAll( element, 'iframe' ).forEach( el => {\n\t\t\t\tif( el.contentWindow ) el.contentWindow.postMessage( 'slide:stop', '*' );\n\t\t\t\tel.removeEventListener( 'load', this.startEmbeddedIframe );\n\t\t\t});\n\n\t\t\t// YouTube postMessage API\n\t\t\tqueryAll( element, 'iframe[src*=\"youtube.com/embed/\"]' ).forEach( el => {\n\t\t\t\tif( !el.hasAttribute( 'data-ignore' ) && el.contentWindow && typeof el.contentWindow.postMessage === 'function' ) {\n\t\t\t\t\tel.contentWindow.postMessage( '{\"event\":\"command\",\"func\":\"pauseVideo\",\"args\":\"\"}', '*' );\n\t\t\t\t}\n\t\t\t});\n\n\t\t\t// Vimeo postMessage API\n\t\t\tqueryAll( element, 'iframe[src*=\"player.vimeo.com/\"]' ).forEach( el => {\n\t\t\t\tif( !el.hasAttribute( 'data-ignore' ) && el.contentWindow && typeof el.contentWindow.postMessage === 'function' ) {\n\t\t\t\t\tel.contentWindow.postMessage( '{\"method\":\"pause\"}', '*' );\n\t\t\t\t}\n\t\t\t});\n\n\t\t\tif( options.unloadIframes === true ) {\n\t\t\t\t// Unload lazy-loaded iframes\n\t\t\t\tqueryAll( element, 'iframe[data-src]' ).forEach( el => {\n\t\t\t\t\t// Only removing the src doesn't actually unload the frame\n\t\t\t\t\t// in all browsers (Firefox) so we set it to blank first\n\t\t\t\t\tel.setAttribute( 'src', 'about:blank' );\n\t\t\t\t\tel.removeAttribute( 'src' );\n\t\t\t\t} );\n\t\t\t}\n\t\t}\n\n\t}\n\n}\n","/**\n * Handles the display of reveal.js' optional slide number.\n */\nexport default class SlideNumber {\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t}\n\n\trender() {\n\n\t\tthis.element = document.createElement( 'div' );\n\t\tthis.element.className = 'slide-number';\n\t\tthis.Reveal.getRevealElement().appendChild( this.element );\n\n\t}\n\n\t/**\n\t * Called when the reveal.js config is updated.\n\t */\n\tconfigure( config, oldConfig ) {\n\n\t\tlet slideNumberDisplay = 'none';\n\t\tif( config.slideNumber && !this.Reveal.isPrintingPDF() ) {\n\t\t\tif( config.showSlideNumber === 'all' ) {\n\t\t\t\tslideNumberDisplay = 'block';\n\t\t\t}\n\t\t\telse if( config.showSlideNumber === 'speaker' && this.Reveal.isSpeakerNotes() ) {\n\t\t\t\tslideNumberDisplay = 'block';\n\t\t\t}\n\t\t}\n\n\t\tthis.element.style.display = slideNumberDisplay;\n\n\t}\n\n\t/**\n\t * Updates the slide number to match the current slide.\n\t */\n\tupdate() {\n\n\t\t// Update slide number if enabled\n\t\tif( this.Reveal.getConfig().slideNumber && this.element ) {\n\t\t\tthis.element.innerHTML = this.getSlideNumber();\n\t\t}\n\n\t}\n\n\t/**\n\t * Returns the HTML string corresponding to the current slide\n\t * number, including formatting.\n\t */\n\tgetSlideNumber( slide = this.Reveal.getCurrentSlide() ) {\n\n\t\tlet config = this.Reveal.getConfig();\n\t\tlet value;\n\t\tlet format = 'h.v';\n\n\t\tif ( typeof config.slideNumber === 'function' ) {\n\t\t\tvalue = config.slideNumber( slide );\n\t\t} else {\n\t\t\t// Check if a custom number format is available\n\t\t\tif( typeof config.slideNumber === 'string' ) {\n\t\t\t\tformat = config.slideNumber;\n\t\t\t}\n\n\t\t\t// If there are ONLY vertical slides in this deck, always use\n\t\t\t// a flattened slide number\n\t\t\tif( !/c/.test( format ) && this.Reveal.getHorizontalSlides().length === 1 ) {\n\t\t\t\tformat = 'c';\n\t\t\t}\n\n\t\t\t// Offset the current slide number by 1 to make it 1-indexed\n\t\t\tlet horizontalOffset = slide && slide.dataset.visibility === 'uncounted' ? 0 : 1;\n\n\t\t\tvalue = [];\n\t\t\tswitch( format ) {\n\t\t\t\tcase 'c':\n\t\t\t\t\tvalue.push( this.Reveal.getSlidePastCount( slide ) + horizontalOffset );\n\t\t\t\t\tbreak;\n\t\t\t\tcase 'c/t':\n\t\t\t\t\tvalue.push( this.Reveal.getSlidePastCount( slide ) + horizontalOffset, '/', this.Reveal.getTotalSlides() );\n\t\t\t\t\tbreak;\n\t\t\t\tdefault:\n\t\t\t\t\tlet indices = this.Reveal.getIndices( slide );\n\t\t\t\t\tvalue.push( indices.h + horizontalOffset );\n\t\t\t\t\tlet sep = format === 'h/v' ? '/' : '.';\n\t\t\t\t\tif( this.Reveal.isVerticalSlide( slide ) ) value.push( sep, indices.v + 1 );\n\t\t\t}\n\t\t}\n\n\t\tlet url = '#' + this.Reveal.location.getHash( slide );\n\t\treturn this.formatNumber( value[0], value[1], value[2], url );\n\n\t}\n\n\t/**\n\t * Applies HTML formatting to a slide number before it's\n\t * written to the DOM.\n\t *\n\t * @param {number} a Current slide\n\t * @param {string} delimiter Character to separate slide numbers\n\t * @param {(number|*)} b Total slides\n\t * @param {HTMLElement} [url='#'+locationHash()] The url to link to\n\t * @return {string} HTML string fragment\n\t */\n\tformatNumber( a, delimiter, b, url = '#' + this.Reveal.location.getHash() ) {\n\n\t\tif( typeof b === 'number' && !isNaN( b ) ) {\n\t\t\treturn `\n\t\t\t\t\t${a}\n\t\t\t\t\t${delimiter}\n\t\t\t\t\t${b}\n\t\t\t\t\t`;\n\t\t}\n\t\telse {\n\t\t\treturn `\n\t\t\t\t\t${a}\n\t\t\t\t\t`;\n\t\t}\n\n\t}\n\n\tdestroy() {\n\n\t\tthis.element.remove();\n\n\t}\n\n}","/**\n * Converts various color input formats to an {r:0,g:0,b:0} object.\n *\n * @param {string} color The string representation of a color\n * @example\n * colorToRgb('#000');\n * @example\n * colorToRgb('#000000');\n * @example\n * colorToRgb('rgb(0,0,0)');\n * @example\n * colorToRgb('rgba(0,0,0)');\n *\n * @return {{r: number, g: number, b: number, [a]: number}|null}\n */\nexport const colorToRgb = ( color ) => {\n\n\tlet hex3 = color.match( /^#([0-9a-f]{3})$/i );\n\tif( hex3 && hex3[1] ) {\n\t\thex3 = hex3[1];\n\t\treturn {\n\t\t\tr: parseInt( hex3.charAt( 0 ), 16 ) * 0x11,\n\t\t\tg: parseInt( hex3.charAt( 1 ), 16 ) * 0x11,\n\t\t\tb: parseInt( hex3.charAt( 2 ), 16 ) * 0x11\n\t\t};\n\t}\n\n\tlet hex6 = color.match( /^#([0-9a-f]{6})$/i );\n\tif( hex6 && hex6[1] ) {\n\t\thex6 = hex6[1];\n\t\treturn {\n\t\t\tr: parseInt( hex6.slice( 0, 2 ), 16 ),\n\t\t\tg: parseInt( hex6.slice( 2, 4 ), 16 ),\n\t\t\tb: parseInt( hex6.slice( 4, 6 ), 16 )\n\t\t};\n\t}\n\n\tlet rgb = color.match( /^rgb\\s*\\(\\s*(\\d+)\\s*,\\s*(\\d+)\\s*,\\s*(\\d+)\\s*\\)$/i );\n\tif( rgb ) {\n\t\treturn {\n\t\t\tr: parseInt( rgb[1], 10 ),\n\t\t\tg: parseInt( rgb[2], 10 ),\n\t\t\tb: parseInt( rgb[3], 10 )\n\t\t};\n\t}\n\n\tlet rgba = color.match( /^rgba\\s*\\(\\s*(\\d+)\\s*,\\s*(\\d+)\\s*,\\s*(\\d+)\\s*\\,\\s*([\\d]+|[\\d]*.[\\d]+)\\s*\\)$/i );\n\tif( rgba ) {\n\t\treturn {\n\t\t\tr: parseInt( rgba[1], 10 ),\n\t\t\tg: parseInt( rgba[2], 10 ),\n\t\t\tb: parseInt( rgba[3], 10 ),\n\t\t\ta: parseFloat( rgba[4] )\n\t\t};\n\t}\n\n\treturn null;\n\n}\n\n/**\n * Calculates brightness on a scale of 0-255.\n *\n * @param {string} color See colorToRgb for supported formats.\n * @see {@link colorToRgb}\n */\nexport const colorBrightness = ( color ) => {\n\n\tif( typeof color === 'string' ) color = colorToRgb( color );\n\n\tif( color ) {\n\t\treturn ( color.r * 299 + color.g * 587 + color.b * 114 ) / 1000;\n\t}\n\n\treturn null;\n\n}","import { queryAll } from '../utils/util.js'\nimport { colorToRgb, colorBrightness } from '../utils/color.js'\n\n/**\n * Creates and updates slide backgrounds.\n */\nexport default class Backgrounds {\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t}\n\n\trender() {\n\n\t\tthis.element = document.createElement( 'div' );\n\t\tthis.element.className = 'backgrounds';\n\t\tthis.Reveal.getRevealElement().appendChild( this.element );\n\n\t}\n\n\t/**\n\t * Creates the slide background elements and appends them\n\t * to the background container. One element is created per\n\t * slide no matter if the given slide has visible background.\n\t */\n\tcreate() {\n\n\t\t// Clear prior backgrounds\n\t\tthis.element.innerHTML = '';\n\t\tthis.element.classList.add( 'no-transition' );\n\n\t\t// Iterate over all horizontal slides\n\t\tthis.Reveal.getHorizontalSlides().forEach( slideh => {\n\n\t\t\tlet backgroundStack = this.createBackground( slideh, this.element );\n\n\t\t\t// Iterate over all vertical slides\n\t\t\tqueryAll( slideh, 'section' ).forEach( slidev => {\n\n\t\t\t\tthis.createBackground( slidev, backgroundStack );\n\n\t\t\t\tbackgroundStack.classList.add( 'stack' );\n\n\t\t\t} );\n\n\t\t} );\n\n\t\t// Add parallax background if specified\n\t\tif( this.Reveal.getConfig().parallaxBackgroundImage ) {\n\n\t\t\tthis.element.style.backgroundImage = 'url(\"' + this.Reveal.getConfig().parallaxBackgroundImage + '\")';\n\t\t\tthis.element.style.backgroundSize = this.Reveal.getConfig().parallaxBackgroundSize;\n\t\t\tthis.element.style.backgroundRepeat = this.Reveal.getConfig().parallaxBackgroundRepeat;\n\t\t\tthis.element.style.backgroundPosition = this.Reveal.getConfig().parallaxBackgroundPosition;\n\n\t\t\t// Make sure the below properties are set on the element - these properties are\n\t\t\t// needed for proper transitions to be set on the element via CSS. To remove\n\t\t\t// annoying background slide-in effect when the presentation starts, apply\n\t\t\t// these properties after short time delay\n\t\t\tsetTimeout( () => {\n\t\t\t\tthis.Reveal.getRevealElement().classList.add( 'has-parallax-background' );\n\t\t\t}, 1 );\n\n\t\t}\n\t\telse {\n\n\t\t\tthis.element.style.backgroundImage = '';\n\t\t\tthis.Reveal.getRevealElement().classList.remove( 'has-parallax-background' );\n\n\t\t}\n\n\t}\n\n\t/**\n\t * Creates a background for the given slide.\n\t *\n\t * @param {HTMLElement} slide\n\t * @param {HTMLElement} container The element that the background\n\t * should be appended to\n\t * @return {HTMLElement} New background div\n\t */\n\tcreateBackground( slide, container ) {\n\n\t\t// Main slide background element\n\t\tlet element = document.createElement( 'div' );\n\t\telement.className = 'slide-background ' + slide.className.replace( /present|past|future/, '' );\n\n\t\t// Inner background element that wraps images/videos/iframes\n\t\tlet contentElement = document.createElement( 'div' );\n\t\tcontentElement.className = 'slide-background-content';\n\n\t\telement.appendChild( contentElement );\n\t\tcontainer.appendChild( element );\n\n\t\tslide.slideBackgroundElement = element;\n\t\tslide.slideBackgroundContentElement = contentElement;\n\n\t\t// Syncs the background to reflect all current background settings\n\t\tthis.sync( slide );\n\n\t\treturn element;\n\n\t}\n\n\t/**\n\t * Renders all of the visual properties of a slide background\n\t * based on the various background attributes.\n\t *\n\t * @param {HTMLElement} slide\n\t */\n\tsync( slide ) {\n\n\t\tconst element = slide.slideBackgroundElement,\n\t\t\tcontentElement = slide.slideBackgroundContentElement;\n\n\t\tconst data = {\n\t\t\tbackground: slide.getAttribute( 'data-background' ),\n\t\t\tbackgroundSize: slide.getAttribute( 'data-background-size' ),\n\t\t\tbackgroundImage: slide.getAttribute( 'data-background-image' ),\n\t\t\tbackgroundVideo: slide.getAttribute( 'data-background-video' ),\n\t\t\tbackgroundIframe: slide.getAttribute( 'data-background-iframe' ),\n\t\t\tbackgroundColor: slide.getAttribute( 'data-background-color' ),\n\t\t\tbackgroundRepeat: slide.getAttribute( 'data-background-repeat' ),\n\t\t\tbackgroundPosition: slide.getAttribute( 'data-background-position' ),\n\t\t\tbackgroundTransition: slide.getAttribute( 'data-background-transition' ),\n\t\t\tbackgroundOpacity: slide.getAttribute( 'data-background-opacity' ),\n\t\t};\n\n\t\tconst dataPreload = slide.hasAttribute( 'data-preload' );\n\n\t\t// Reset the prior background state in case this is not the\n\t\t// initial sync\n\t\tslide.classList.remove( 'has-dark-background' );\n\t\tslide.classList.remove( 'has-light-background' );\n\n\t\telement.removeAttribute( 'data-loaded' );\n\t\telement.removeAttribute( 'data-background-hash' );\n\t\telement.removeAttribute( 'data-background-size' );\n\t\telement.removeAttribute( 'data-background-transition' );\n\t\telement.style.backgroundColor = '';\n\n\t\tcontentElement.style.backgroundSize = '';\n\t\tcontentElement.style.backgroundRepeat = '';\n\t\tcontentElement.style.backgroundPosition = '';\n\t\tcontentElement.style.backgroundImage = '';\n\t\tcontentElement.style.opacity = '';\n\t\tcontentElement.innerHTML = '';\n\n\t\tif( data.background ) {\n\t\t\t// Auto-wrap image urls in url(...)\n\t\t\tif( /^(http|file|\\/\\/)/gi.test( data.background ) || /\\.(svg|png|jpg|jpeg|gif|bmp)([?#\\s]|$)/gi.test( data.background ) ) {\n\t\t\t\tslide.setAttribute( 'data-background-image', data.background );\n\t\t\t}\n\t\t\telse {\n\t\t\t\telement.style.background = data.background;\n\t\t\t}\n\t\t}\n\n\t\t// Create a hash for this combination of background settings.\n\t\t// This is used to determine when two slide backgrounds are\n\t\t// the same.\n\t\tif( data.background || data.backgroundColor || data.backgroundImage || data.backgroundVideo || data.backgroundIframe ) {\n\t\t\telement.setAttribute( 'data-background-hash', data.background +\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\tdata.backgroundSize +\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\tdata.backgroundImage +\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\tdata.backgroundVideo +\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\tdata.backgroundIframe +\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\tdata.backgroundColor +\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\tdata.backgroundRepeat +\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\tdata.backgroundPosition +\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\tdata.backgroundTransition +\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\tdata.backgroundOpacity );\n\t\t}\n\n\t\t// Additional and optional background properties\n\t\tif( data.backgroundSize ) element.setAttribute( 'data-background-size', data.backgroundSize );\n\t\tif( data.backgroundColor ) element.style.backgroundColor = data.backgroundColor;\n\t\tif( data.backgroundTransition ) element.setAttribute( 'data-background-transition', data.backgroundTransition );\n\n\t\tif( dataPreload ) element.setAttribute( 'data-preload', '' );\n\n\t\t// Background image options are set on the content wrapper\n\t\tif( data.backgroundSize ) contentElement.style.backgroundSize = data.backgroundSize;\n\t\tif( data.backgroundRepeat ) contentElement.style.backgroundRepeat = data.backgroundRepeat;\n\t\tif( data.backgroundPosition ) contentElement.style.backgroundPosition = data.backgroundPosition;\n\t\tif( data.backgroundOpacity ) contentElement.style.opacity = data.backgroundOpacity;\n\n\t\t// If this slide has a background color, we add a class that\n\t\t// signals if it is light or dark. If the slide has no background\n\t\t// color, no class will be added\n\t\tlet contrastColor = data.backgroundColor;\n\n\t\t// If no bg color was found, or it cannot be converted by colorToRgb, check the computed background\n\t\tif( !contrastColor || !colorToRgb( contrastColor ) ) {\n\t\t\tlet computedBackgroundStyle = window.getComputedStyle( element );\n\t\t\tif( computedBackgroundStyle && computedBackgroundStyle.backgroundColor ) {\n\t\t\t\tcontrastColor = computedBackgroundStyle.backgroundColor;\n\t\t\t}\n\t\t}\n\n\t\tif( contrastColor ) {\n\t\t\tconst rgb = colorToRgb( contrastColor );\n\n\t\t\t// Ignore fully transparent backgrounds. Some browsers return\n\t\t\t// rgba(0,0,0,0) when reading the computed background color of\n\t\t\t// an element with no background\n\t\t\tif( rgb && rgb.a !== 0 ) {\n\t\t\t\tif( colorBrightness( contrastColor ) < 128 ) {\n\t\t\t\t\tslide.classList.add( 'has-dark-background' );\n\t\t\t\t}\n\t\t\t\telse {\n\t\t\t\t\tslide.classList.add( 'has-light-background' );\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t}\n\n\t/**\n\t * Updates the background elements to reflect the current\n\t * slide.\n\t *\n\t * @param {boolean} includeAll If true, the backgrounds of\n\t * all vertical slides (not just the present) will be updated.\n\t */\n\tupdate( includeAll = false ) {\n\n\t\tlet currentSlide = this.Reveal.getCurrentSlide();\n\t\tlet indices = this.Reveal.getIndices();\n\n\t\tlet currentBackground = null;\n\n\t\t// Reverse past/future classes when in RTL mode\n\t\tlet horizontalPast = this.Reveal.getConfig().rtl ? 'future' : 'past',\n\t\t\thorizontalFuture = this.Reveal.getConfig().rtl ? 'past' : 'future';\n\n\t\t// Update the classes of all backgrounds to match the\n\t\t// states of their slides (past/present/future)\n\t\tArray.from( this.element.childNodes ).forEach( ( backgroundh, h ) => {\n\n\t\t\tbackgroundh.classList.remove( 'past', 'present', 'future' );\n\n\t\t\tif( h < indices.h ) {\n\t\t\t\tbackgroundh.classList.add( horizontalPast );\n\t\t\t}\n\t\t\telse if ( h > indices.h ) {\n\t\t\t\tbackgroundh.classList.add( horizontalFuture );\n\t\t\t}\n\t\t\telse {\n\t\t\t\tbackgroundh.classList.add( 'present' );\n\n\t\t\t\t// Store a reference to the current background element\n\t\t\t\tcurrentBackground = backgroundh;\n\t\t\t}\n\n\t\t\tif( includeAll || h === indices.h ) {\n\t\t\t\tqueryAll( backgroundh, '.slide-background' ).forEach( ( backgroundv, v ) => {\n\n\t\t\t\t\tbackgroundv.classList.remove( 'past', 'present', 'future' );\n\n\t\t\t\t\tif( v < indices.v ) {\n\t\t\t\t\t\tbackgroundv.classList.add( 'past' );\n\t\t\t\t\t}\n\t\t\t\t\telse if ( v > indices.v ) {\n\t\t\t\t\t\tbackgroundv.classList.add( 'future' );\n\t\t\t\t\t}\n\t\t\t\t\telse {\n\t\t\t\t\t\tbackgroundv.classList.add( 'present' );\n\n\t\t\t\t\t\t// Only if this is the present horizontal and vertical slide\n\t\t\t\t\t\tif( h === indices.h ) currentBackground = backgroundv;\n\t\t\t\t\t}\n\n\t\t\t\t} );\n\t\t\t}\n\n\t\t} );\n\n\t\t// Stop content inside of previous backgrounds\n\t\tif( this.previousBackground ) {\n\n\t\t\tthis.Reveal.slideContent.stopEmbeddedContent( this.previousBackground, { unloadIframes: !this.Reveal.slideContent.shouldPreload( this.previousBackground ) } );\n\n\t\t}\n\n\t\t// Start content in the current background\n\t\tif( currentBackground ) {\n\n\t\t\tthis.Reveal.slideContent.startEmbeddedContent( currentBackground );\n\n\t\t\tlet currentBackgroundContent = currentBackground.querySelector( '.slide-background-content' );\n\t\t\tif( currentBackgroundContent ) {\n\n\t\t\t\tlet backgroundImageURL = currentBackgroundContent.style.backgroundImage || '';\n\n\t\t\t\t// Restart GIFs (doesn't work in Firefox)\n\t\t\t\tif( /\\.gif/i.test( backgroundImageURL ) ) {\n\t\t\t\t\tcurrentBackgroundContent.style.backgroundImage = '';\n\t\t\t\t\twindow.getComputedStyle( currentBackgroundContent ).opacity;\n\t\t\t\t\tcurrentBackgroundContent.style.backgroundImage = backgroundImageURL;\n\t\t\t\t}\n\n\t\t\t}\n\n\t\t\t// Don't transition between identical backgrounds. This\n\t\t\t// prevents unwanted flicker.\n\t\t\tlet previousBackgroundHash = this.previousBackground ? this.previousBackground.getAttribute( 'data-background-hash' ) : null;\n\t\t\tlet currentBackgroundHash = currentBackground.getAttribute( 'data-background-hash' );\n\t\t\tif( currentBackgroundHash && currentBackgroundHash === previousBackgroundHash && currentBackground !== this.previousBackground ) {\n\t\t\t\tthis.element.classList.add( 'no-transition' );\n\t\t\t}\n\n\t\t\tthis.previousBackground = currentBackground;\n\n\t\t}\n\n\t\t// If there's a background brightness flag for this slide,\n\t\t// bubble it to the .reveal container\n\t\tif( currentSlide ) {\n\t\t\t[ 'has-light-background', 'has-dark-background' ].forEach( classToBubble => {\n\t\t\t\tif( currentSlide.classList.contains( classToBubble ) ) {\n\t\t\t\t\tthis.Reveal.getRevealElement().classList.add( classToBubble );\n\t\t\t\t}\n\t\t\t\telse {\n\t\t\t\t\tthis.Reveal.getRevealElement().classList.remove( classToBubble );\n\t\t\t\t}\n\t\t\t}, this );\n\t\t}\n\n\t\t// Allow the first background to apply without transition\n\t\tsetTimeout( () => {\n\t\t\tthis.element.classList.remove( 'no-transition' );\n\t\t}, 1 );\n\n\t}\n\n\t/**\n\t * Updates the position of the parallax background based\n\t * on the current slide index.\n\t */\n\tupdateParallax() {\n\n\t\tlet indices = this.Reveal.getIndices();\n\n\t\tif( this.Reveal.getConfig().parallaxBackgroundImage ) {\n\n\t\t\tlet horizontalSlides = this.Reveal.getHorizontalSlides(),\n\t\t\t\tverticalSlides = this.Reveal.getVerticalSlides();\n\n\t\t\tlet backgroundSize = this.element.style.backgroundSize.split( ' ' ),\n\t\t\t\tbackgroundWidth, backgroundHeight;\n\n\t\t\tif( backgroundSize.length === 1 ) {\n\t\t\t\tbackgroundWidth = backgroundHeight = parseInt( backgroundSize[0], 10 );\n\t\t\t}\n\t\t\telse {\n\t\t\t\tbackgroundWidth = parseInt( backgroundSize[0], 10 );\n\t\t\t\tbackgroundHeight = parseInt( backgroundSize[1], 10 );\n\t\t\t}\n\n\t\t\tlet slideWidth = this.element.offsetWidth,\n\t\t\t\thorizontalSlideCount = horizontalSlides.length,\n\t\t\t\thorizontalOffsetMultiplier,\n\t\t\t\thorizontalOffset;\n\n\t\t\tif( typeof this.Reveal.getConfig().parallaxBackgroundHorizontal === 'number' ) {\n\t\t\t\thorizontalOffsetMultiplier = this.Reveal.getConfig().parallaxBackgroundHorizontal;\n\t\t\t}\n\t\t\telse {\n\t\t\t\thorizontalOffsetMultiplier = horizontalSlideCount > 1 ? ( backgroundWidth - slideWidth ) / ( horizontalSlideCount-1 ) : 0;\n\t\t\t}\n\n\t\t\thorizontalOffset = horizontalOffsetMultiplier * indices.h * -1;\n\n\t\t\tlet slideHeight = this.element.offsetHeight,\n\t\t\t\tverticalSlideCount = verticalSlides.length,\n\t\t\t\tverticalOffsetMultiplier,\n\t\t\t\tverticalOffset;\n\n\t\t\tif( typeof this.Reveal.getConfig().parallaxBackgroundVertical === 'number' ) {\n\t\t\t\tverticalOffsetMultiplier = this.Reveal.getConfig().parallaxBackgroundVertical;\n\t\t\t}\n\t\t\telse {\n\t\t\t\tverticalOffsetMultiplier = ( backgroundHeight - slideHeight ) / ( verticalSlideCount-1 );\n\t\t\t}\n\n\t\t\tverticalOffset = verticalSlideCount > 0 ? verticalOffsetMultiplier * indices.v : 0;\n\n\t\t\tthis.element.style.backgroundPosition = horizontalOffset + 'px ' + -verticalOffset + 'px';\n\n\t\t}\n\n\t}\n\n\tdestroy() {\n\n\t\tthis.element.remove();\n\n\t}\n\n}\n","\nexport const SLIDES_SELECTOR = '.slides section';\nexport const HORIZONTAL_SLIDES_SELECTOR = '.slides>section';\nexport const VERTICAL_SLIDES_SELECTOR = '.slides>section.present>section';\n\n// Methods that may not be invoked via the postMessage API\nexport const POST_MESSAGE_METHOD_BLACKLIST = /registerPlugin|registerKeyboardShortcut|addKeyBinding|addEventListener/;\n\n// Regex for retrieving the fragment style from a class attribute\nexport const FRAGMENT_STYLE_REGEX = /fade-(down|up|right|left|out|in-then-out|in-then-semi-out)|semi-fade-out|current-visible|shrink|grow/;","import { queryAll, extend, createStyleSheet, matches, closest } from '../utils/util.js'\nimport { FRAGMENT_STYLE_REGEX } from '../utils/constants.js'\n\n// Counter used to generate unique IDs for auto-animated elements\nlet autoAnimateCounter = 0;\n\n/**\n * Automatically animates matching elements across\n * slides with the [data-auto-animate] attribute.\n */\nexport default class AutoAnimate {\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t}\n\n\t/**\n\t * Runs an auto-animation between the given slides.\n\t *\n\t * @param {HTMLElement} fromSlide\n\t * @param {HTMLElement} toSlide\n\t */\n\trun( fromSlide, toSlide ) {\n\n\t\t// Clean up after prior animations\n\t\tthis.reset();\n\n\t\tlet allSlides = this.Reveal.getSlides();\n\t\tlet toSlideIndex = allSlides.indexOf( toSlide );\n\t\tlet fromSlideIndex = allSlides.indexOf( fromSlide );\n\n\t\t// Ensure that both slides are auto-animate targets with the same data-auto-animate-id value\n\t\t// (including null if absent on both) and that data-auto-animate-restart isn't set on the\n\t\t// physically latter slide (independent of slide direction)\n\t\tif( fromSlide.hasAttribute( 'data-auto-animate' ) && toSlide.hasAttribute( 'data-auto-animate' )\n\t\t\t\t&& fromSlide.getAttribute( 'data-auto-animate-id' ) === toSlide.getAttribute( 'data-auto-animate-id' ) \n\t\t\t\t&& !( toSlideIndex > fromSlideIndex ? toSlide : fromSlide ).hasAttribute( 'data-auto-animate-restart' ) ) {\n\n\t\t\t// Create a new auto-animate sheet\n\t\t\tthis.autoAnimateStyleSheet = this.autoAnimateStyleSheet || createStyleSheet();\n\n\t\t\tlet animationOptions = this.getAutoAnimateOptions( toSlide );\n\n\t\t\t// Set our starting state\n\t\t\tfromSlide.dataset.autoAnimate = 'pending';\n\t\t\ttoSlide.dataset.autoAnimate = 'pending';\n\n\t\t\t// Flag the navigation direction, needed for fragment buildup\n\t\t\tanimationOptions.slideDirection = toSlideIndex > fromSlideIndex ? 'forward' : 'backward';\n\n\t\t\t// Inject our auto-animate styles for this transition\n\t\t\tlet css = this.getAutoAnimatableElements( fromSlide, toSlide ).map( elements => {\n\t\t\t\treturn this.autoAnimateElements( elements.from, elements.to, elements.options || {}, animationOptions, autoAnimateCounter++ );\n\t\t\t} );\n\n\t\t\t// Animate unmatched elements, if enabled\n\t\t\tif( toSlide.dataset.autoAnimateUnmatched !== 'false' && this.Reveal.getConfig().autoAnimateUnmatched === true ) {\n\n\t\t\t\t// Our default timings for unmatched elements\n\t\t\t\tlet defaultUnmatchedDuration = animationOptions.duration * 0.8,\n\t\t\t\t\tdefaultUnmatchedDelay = animationOptions.duration * 0.2;\n\n\t\t\t\tthis.getUnmatchedAutoAnimateElements( toSlide ).forEach( unmatchedElement => {\n\n\t\t\t\t\tlet unmatchedOptions = this.getAutoAnimateOptions( unmatchedElement, animationOptions );\n\t\t\t\t\tlet id = 'unmatched';\n\n\t\t\t\t\t// If there is a duration or delay set specifically for this\n\t\t\t\t\t// element our unmatched elements should adhere to those\n\t\t\t\t\tif( unmatchedOptions.duration !== animationOptions.duration || unmatchedOptions.delay !== animationOptions.delay ) {\n\t\t\t\t\t\tid = 'unmatched-' + autoAnimateCounter++;\n\t\t\t\t\t\tcss.push( `[data-auto-animate=\"running\"] [data-auto-animate-target=\"${id}\"] { transition: opacity ${unmatchedOptions.duration}s ease ${unmatchedOptions.delay}s; }` );\n\t\t\t\t\t}\n\n\t\t\t\t\tunmatchedElement.dataset.autoAnimateTarget = id;\n\n\t\t\t\t}, this );\n\n\t\t\t\t// Our default transition for unmatched elements\n\t\t\t\tcss.push( `[data-auto-animate=\"running\"] [data-auto-animate-target=\"unmatched\"] { transition: opacity ${defaultUnmatchedDuration}s ease ${defaultUnmatchedDelay}s; }` );\n\n\t\t\t}\n\n\t\t\t// Setting the whole chunk of CSS at once is the most\n\t\t\t// efficient way to do this. Using sheet.insertRule\n\t\t\t// is multiple factors slower.\n\t\t\tthis.autoAnimateStyleSheet.innerHTML = css.join( '' );\n\n\t\t\t// Start the animation next cycle\n\t\t\trequestAnimationFrame( () => {\n\t\t\t\tif( this.autoAnimateStyleSheet ) {\n\t\t\t\t\t// This forces our newly injected styles to be applied in Firefox\n\t\t\t\t\tgetComputedStyle( this.autoAnimateStyleSheet ).fontWeight;\n\n\t\t\t\t\ttoSlide.dataset.autoAnimate = 'running';\n\t\t\t\t}\n\t\t\t} );\n\n\t\t\tthis.Reveal.dispatchEvent({\n\t\t\t\ttype: 'autoanimate',\n\t\t\t\tdata: {\n\t\t\t\t\tfromSlide,\n\t\t\t\t\ttoSlide,\n\t\t\t\t\tsheet: this.autoAnimateStyleSheet\n\t\t\t\t}\n\t\t\t});\n\n\t\t}\n\n\t}\n\n\t/**\n\t * Rolls back all changes that we've made to the DOM so\n\t * that as part of animating.\n\t */\n\treset() {\n\n\t\t// Reset slides\n\t\tqueryAll( this.Reveal.getRevealElement(), '[data-auto-animate]:not([data-auto-animate=\"\"])' ).forEach( element => {\n\t\t\telement.dataset.autoAnimate = '';\n\t\t} );\n\n\t\t// Reset elements\n\t\tqueryAll( this.Reveal.getRevealElement(), '[data-auto-animate-target]' ).forEach( element => {\n\t\t\tdelete element.dataset.autoAnimateTarget;\n\t\t} );\n\n\t\t// Remove the animation sheet\n\t\tif( this.autoAnimateStyleSheet && this.autoAnimateStyleSheet.parentNode ) {\n\t\t\tthis.autoAnimateStyleSheet.parentNode.removeChild( this.autoAnimateStyleSheet );\n\t\t\tthis.autoAnimateStyleSheet = null;\n\t\t}\n\n\t}\n\n\t/**\n\t * Creates a FLIP animation where the `to` element starts out\n\t * in the `from` element position and animates to its original\n\t * state.\n\t *\n\t * @param {HTMLElement} from\n\t * @param {HTMLElement} to\n\t * @param {Object} elementOptions Options for this element pair\n\t * @param {Object} animationOptions Options set at the slide level\n\t * @param {String} id Unique ID that we can use to identify this\n\t * auto-animate element in the DOM\n\t */\n\tautoAnimateElements( from, to, elementOptions, animationOptions, id ) {\n\n\t\t// 'from' elements are given a data-auto-animate-target with no value,\n\t\t// 'to' elements are are given a data-auto-animate-target with an ID\n\t\tfrom.dataset.autoAnimateTarget = '';\n\t\tto.dataset.autoAnimateTarget = id;\n\n\t\t// Each element may override any of the auto-animate options\n\t\t// like transition easing, duration and delay via data-attributes\n\t\tlet options = this.getAutoAnimateOptions( to, animationOptions );\n\n\t\t// If we're using a custom element matcher the element options\n\t\t// may contain additional transition overrides\n\t\tif( typeof elementOptions.delay !== 'undefined' ) options.delay = elementOptions.delay;\n\t\tif( typeof elementOptions.duration !== 'undefined' ) options.duration = elementOptions.duration;\n\t\tif( typeof elementOptions.easing !== 'undefined' ) options.easing = elementOptions.easing;\n\n\t\tlet fromProps = this.getAutoAnimatableProperties( 'from', from, elementOptions ),\n\t\t\ttoProps = this.getAutoAnimatableProperties( 'to', to, elementOptions );\n\n\t\t// Maintain fragment visibility for matching elements when\n\t\t// we're navigating forwards, this way the viewer won't need\n\t\t// to step through the same fragments twice\n\t\tif( to.classList.contains( 'fragment' ) ) {\n\n\t\t\t// Don't auto-animate the opacity of fragments to avoid\n\t\t\t// conflicts with fragment animations\n\t\t\tdelete toProps.styles['opacity'];\n\n\t\t\tif( from.classList.contains( 'fragment' ) ) {\n\n\t\t\t\tlet fromFragmentStyle = ( from.className.match( FRAGMENT_STYLE_REGEX ) || [''] )[0];\n\t\t\t\tlet toFragmentStyle = ( to.className.match( FRAGMENT_STYLE_REGEX ) || [''] )[0];\n\n\t\t\t\t// Only skip the fragment if the fragment animation style\n\t\t\t\t// remains unchanged\n\t\t\t\tif( fromFragmentStyle === toFragmentStyle && animationOptions.slideDirection === 'forward' ) {\n\t\t\t\t\tto.classList.add( 'visible', 'disabled' );\n\t\t\t\t}\n\n\t\t\t}\n\n\t\t}\n\n\t\t// If translation and/or scaling are enabled, css transform\n\t\t// the 'to' element so that it matches the position and size\n\t\t// of the 'from' element\n\t\tif( elementOptions.translate !== false || elementOptions.scale !== false ) {\n\n\t\t\tlet presentationScale = this.Reveal.getScale();\n\n\t\t\tlet delta = {\n\t\t\t\tx: ( fromProps.x - toProps.x ) / presentationScale,\n\t\t\t\ty: ( fromProps.y - toProps.y ) / presentationScale,\n\t\t\t\tscaleX: fromProps.width / toProps.width,\n\t\t\t\tscaleY: fromProps.height / toProps.height\n\t\t\t};\n\n\t\t\t// Limit decimal points to avoid 0.0001px blur and stutter\n\t\t\tdelta.x = Math.round( delta.x * 1000 ) / 1000;\n\t\t\tdelta.y = Math.round( delta.y * 1000 ) / 1000;\n\t\t\tdelta.scaleX = Math.round( delta.scaleX * 1000 ) / 1000;\n\t\t\tdelta.scaleX = Math.round( delta.scaleX * 1000 ) / 1000;\n\n\t\t\tlet translate = elementOptions.translate !== false && ( delta.x !== 0 || delta.y !== 0 ),\n\t\t\t\tscale = elementOptions.scale !== false && ( delta.scaleX !== 0 || delta.scaleY !== 0 );\n\n\t\t\t// No need to transform if nothing's changed\n\t\t\tif( translate || scale ) {\n\n\t\t\t\tlet transform = [];\n\n\t\t\t\tif( translate ) transform.push( `translate(${delta.x}px, ${delta.y}px)` );\n\t\t\t\tif( scale ) transform.push( `scale(${delta.scaleX}, ${delta.scaleY})` );\n\n\t\t\t\tfromProps.styles['transform'] = transform.join( ' ' );\n\t\t\t\tfromProps.styles['transform-origin'] = 'top left';\n\n\t\t\t\ttoProps.styles['transform'] = 'none';\n\n\t\t\t}\n\n\t\t}\n\n\t\t// Delete all unchanged 'to' styles\n\t\tfor( let propertyName in toProps.styles ) {\n\t\t\tconst toValue = toProps.styles[propertyName];\n\t\t\tconst fromValue = fromProps.styles[propertyName];\n\n\t\t\tif( toValue === fromValue ) {\n\t\t\t\tdelete toProps.styles[propertyName];\n\t\t\t}\n\t\t\telse {\n\t\t\t\t// If these property values were set via a custom matcher providing\n\t\t\t\t// an explicit 'from' and/or 'to' value, we always inject those values.\n\t\t\t\tif( toValue.explicitValue === true ) {\n\t\t\t\t\ttoProps.styles[propertyName] = toValue.value;\n\t\t\t\t}\n\n\t\t\t\tif( fromValue.explicitValue === true ) {\n\t\t\t\t\tfromProps.styles[propertyName] = fromValue.value;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tlet css = '';\n\n\t\tlet toStyleProperties = Object.keys( toProps.styles );\n\n\t\t// Only create animate this element IF at least one style\n\t\t// property has changed\n\t\tif( toStyleProperties.length > 0 ) {\n\n\t\t\t// Instantly move to the 'from' state\n\t\t\tfromProps.styles['transition'] = 'none';\n\n\t\t\t// Animate towards the 'to' state\n\t\t\ttoProps.styles['transition'] = `all ${options.duration}s ${options.easing} ${options.delay}s`;\n\t\t\ttoProps.styles['transition-property'] = toStyleProperties.join( ', ' );\n\t\t\ttoProps.styles['will-change'] = toStyleProperties.join( ', ' );\n\n\t\t\t// Build up our custom CSS. We need to override inline styles\n\t\t\t// so we need to make our styles vErY IMPORTANT!1!!\n\t\t\tlet fromCSS = Object.keys( fromProps.styles ).map( propertyName => {\n\t\t\t\treturn propertyName + ': ' + fromProps.styles[propertyName] + ' !important;';\n\t\t\t} ).join( '' );\n\n\t\t\tlet toCSS = Object.keys( toProps.styles ).map( propertyName => {\n\t\t\t\treturn propertyName + ': ' + toProps.styles[propertyName] + ' !important;';\n\t\t\t} ).join( '' );\n\n\t\t\tcss = \t'[data-auto-animate-target=\"'+ id +'\"] {'+ fromCSS +'}' +\n\t\t\t\t\t'[data-auto-animate=\"running\"] [data-auto-animate-target=\"'+ id +'\"] {'+ toCSS +'}';\n\n\t\t}\n\n\t\treturn css;\n\n\t}\n\n\t/**\n\t * Returns the auto-animate options for the given element.\n\t *\n\t * @param {HTMLElement} element Element to pick up options\n\t * from, either a slide or an animation target\n\t * @param {Object} [inheritedOptions] Optional set of existing\n\t * options\n\t */\n\tgetAutoAnimateOptions( element, inheritedOptions ) {\n\n\t\tlet options = {\n\t\t\teasing: this.Reveal.getConfig().autoAnimateEasing,\n\t\t\tduration: this.Reveal.getConfig().autoAnimateDuration,\n\t\t\tdelay: 0\n\t\t};\n\n\t\toptions = extend( options, inheritedOptions );\n\n\t\t// Inherit options from parent elements\n\t\tif( element.parentNode ) {\n\t\t\tlet autoAnimatedParent = closest( element.parentNode, '[data-auto-animate-target]' );\n\t\t\tif( autoAnimatedParent ) {\n\t\t\t\toptions = this.getAutoAnimateOptions( autoAnimatedParent, options );\n\t\t\t}\n\t\t}\n\n\t\tif( element.dataset.autoAnimateEasing ) {\n\t\t\toptions.easing = element.dataset.autoAnimateEasing;\n\t\t}\n\n\t\tif( element.dataset.autoAnimateDuration ) {\n\t\t\toptions.duration = parseFloat( element.dataset.autoAnimateDuration );\n\t\t}\n\n\t\tif( element.dataset.autoAnimateDelay ) {\n\t\t\toptions.delay = parseFloat( element.dataset.autoAnimateDelay );\n\t\t}\n\n\t\treturn options;\n\n\t}\n\n\t/**\n\t * Returns an object containing all of the properties\n\t * that can be auto-animated for the given element and\n\t * their current computed values.\n\t *\n\t * @param {String} direction 'from' or 'to'\n\t */\n\tgetAutoAnimatableProperties( direction, element, elementOptions ) {\n\n\t\tlet config = this.Reveal.getConfig();\n\n\t\tlet properties = { styles: [] };\n\n\t\t// Position and size\n\t\tif( elementOptions.translate !== false || elementOptions.scale !== false ) {\n\t\t\tlet bounds;\n\n\t\t\t// Custom auto-animate may optionally return a custom tailored\n\t\t\t// measurement function\n\t\t\tif( typeof elementOptions.measure === 'function' ) {\n\t\t\t\tbounds = elementOptions.measure( element );\n\t\t\t}\n\t\t\telse {\n\t\t\t\tif( config.center ) {\n\t\t\t\t\t// More precise, but breaks when used in combination\n\t\t\t\t\t// with zoom for scaling the deck ¯\\_(ツ)_/¯\n\t\t\t\t\tbounds = element.getBoundingClientRect();\n\t\t\t\t}\n\t\t\t\telse {\n\t\t\t\t\tlet scale = this.Reveal.getScale();\n\t\t\t\t\tbounds = {\n\t\t\t\t\t\tx: element.offsetLeft * scale,\n\t\t\t\t\t\ty: element.offsetTop * scale,\n\t\t\t\t\t\twidth: element.offsetWidth * scale,\n\t\t\t\t\t\theight: element.offsetHeight * scale\n\t\t\t\t\t};\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tproperties.x = bounds.x;\n\t\t\tproperties.y = bounds.y;\n\t\t\tproperties.width = bounds.width;\n\t\t\tproperties.height = bounds.height;\n\t\t}\n\n\t\tconst computedStyles = getComputedStyle( element );\n\n\t\t// CSS styles\n\t\t( elementOptions.styles || config.autoAnimateStyles ).forEach( style => {\n\t\t\tlet value;\n\n\t\t\t// `style` is either the property name directly, or an object\n\t\t\t// definition of a style property\n\t\t\tif( typeof style === 'string' ) style = { property: style };\n\n\t\t\tif( typeof style.from !== 'undefined' && direction === 'from' ) {\n\t\t\t\tvalue = { value: style.from, explicitValue: true };\n\t\t\t}\n\t\t\telse if( typeof style.to !== 'undefined' && direction === 'to' ) {\n\t\t\t\tvalue = { value: style.to, explicitValue: true };\n\t\t\t}\n\t\t\telse {\n\t\t\t\tvalue = computedStyles[style.property];\n\t\t\t}\n\n\t\t\tif( value !== '' ) {\n\t\t\t\tproperties.styles[style.property] = value;\n\t\t\t}\n\t\t} );\n\n\t\treturn properties;\n\n\t}\n\n\t/**\n\t * Get a list of all element pairs that we can animate\n\t * between the given slides.\n\t *\n\t * @param {HTMLElement} fromSlide\n\t * @param {HTMLElement} toSlide\n\t *\n\t * @return {Array} Each value is an array where [0] is\n\t * the element we're animating from and [1] is the\n\t * element we're animating to\n\t */\n\tgetAutoAnimatableElements( fromSlide, toSlide ) {\n\n\t\tlet matcher = typeof this.Reveal.getConfig().autoAnimateMatcher === 'function' ? this.Reveal.getConfig().autoAnimateMatcher : this.getAutoAnimatePairs;\n\n\t\tlet pairs = matcher.call( this, fromSlide, toSlide );\n\n\t\tlet reserved = [];\n\n\t\t// Remove duplicate pairs\n\t\treturn pairs.filter( ( pair, index ) => {\n\t\t\tif( reserved.indexOf( pair.to ) === -1 ) {\n\t\t\t\treserved.push( pair.to );\n\t\t\t\treturn true;\n\t\t\t}\n\t\t} );\n\n\t}\n\n\t/**\n\t * Identifies matching elements between slides.\n\t *\n\t * You can specify a custom matcher function by using\n\t * the `autoAnimateMatcher` config option.\n\t */\n\tgetAutoAnimatePairs( fromSlide, toSlide ) {\n\n\t\tlet pairs = [];\n\n\t\tconst codeNodes = 'pre';\n\t\tconst textNodes = 'h1, h2, h3, h4, h5, h6, p, li';\n\t\tconst mediaNodes = 'img, video, iframe';\n\n\t\t// Eplicit matches via data-id\n\t\tthis.findAutoAnimateMatches( pairs, fromSlide, toSlide, '[data-id]', node => {\n\t\t\treturn node.nodeName + ':::' + node.getAttribute( 'data-id' );\n\t\t} );\n\n\t\t// Text\n\t\tthis.findAutoAnimateMatches( pairs, fromSlide, toSlide, textNodes, node => {\n\t\t\treturn node.nodeName + ':::' + node.innerText;\n\t\t} );\n\n\t\t// Media\n\t\tthis.findAutoAnimateMatches( pairs, fromSlide, toSlide, mediaNodes, node => {\n\t\t\treturn node.nodeName + ':::' + ( node.getAttribute( 'src' ) || node.getAttribute( 'data-src' ) );\n\t\t} );\n\n\t\t// Code\n\t\tthis.findAutoAnimateMatches( pairs, fromSlide, toSlide, codeNodes, node => {\n\t\t\treturn node.nodeName + ':::' + node.innerText;\n\t\t} );\n\n\t\tpairs.forEach( pair => {\n\n\t\t\t// Disable scale transformations on text nodes, we transition\n\t\t\t// each individual text property instead\n\t\t\tif( matches( pair.from, textNodes ) ) {\n\t\t\t\tpair.options = { scale: false };\n\t\t\t}\n\t\t\t// Animate individual lines of code\n\t\t\telse if( matches( pair.from, codeNodes ) ) {\n\n\t\t\t\t// Transition the code block's width and height instead of scaling\n\t\t\t\t// to prevent its content from being squished\n\t\t\t\tpair.options = { scale: false, styles: [ 'width', 'height' ] };\n\n\t\t\t\t// Lines of code\n\t\t\t\tthis.findAutoAnimateMatches( pairs, pair.from, pair.to, '.hljs .hljs-ln-code', node => {\n\t\t\t\t\treturn node.textContent;\n\t\t\t\t}, {\n\t\t\t\t\tscale: false,\n\t\t\t\t\tstyles: [],\n\t\t\t\t\tmeasure: this.getLocalBoundingBox.bind( this )\n\t\t\t\t} );\n\n\t\t\t\t// Line numbers\n\t\t\t\tthis.findAutoAnimateMatches( pairs, pair.from, pair.to, '.hljs .hljs-ln-line[data-line-number]', node => {\n\t\t\t\t\treturn node.getAttribute( 'data-line-number' );\n\t\t\t\t}, {\n\t\t\t\t\tscale: false,\n\t\t\t\t\tstyles: [ 'width' ],\n\t\t\t\t\tmeasure: this.getLocalBoundingBox.bind( this )\n\t\t\t\t} );\n\n\t\t\t}\n\n\t\t}, this );\n\n\t\treturn pairs;\n\n\t}\n\n\t/**\n\t * Helper method which returns a bounding box based on\n\t * the given elements offset coordinates.\n\t *\n\t * @param {HTMLElement} element\n\t * @return {Object} x, y, width, height\n\t */\n\tgetLocalBoundingBox( element ) {\n\n\t\tconst presentationScale = this.Reveal.getScale();\n\n\t\treturn {\n\t\t\tx: Math.round( ( element.offsetLeft * presentationScale ) * 100 ) / 100,\n\t\t\ty: Math.round( ( element.offsetTop * presentationScale ) * 100 ) / 100,\n\t\t\twidth: Math.round( ( element.offsetWidth * presentationScale ) * 100 ) / 100,\n\t\t\theight: Math.round( ( element.offsetHeight * presentationScale ) * 100 ) / 100\n\t\t};\n\n\t}\n\n\t/**\n\t * Finds matching elements between two slides.\n\t *\n\t * @param {Array} pairs \tList of pairs to push matches to\n\t * @param {HTMLElement} fromScope Scope within the from element exists\n\t * @param {HTMLElement} toScope Scope within the to element exists\n\t * @param {String} selector CSS selector of the element to match\n\t * @param {Function} serializer A function that accepts an element and returns\n\t * a stringified ID based on its contents\n\t * @param {Object} animationOptions Optional config options for this pair\n\t */\n\tfindAutoAnimateMatches( pairs, fromScope, toScope, selector, serializer, animationOptions ) {\n\n\t\tlet fromMatches = {};\n\t\tlet toMatches = {};\n\n\t\t[].slice.call( fromScope.querySelectorAll( selector ) ).forEach( ( element, i ) => {\n\t\t\tconst key = serializer( element );\n\t\t\tif( typeof key === 'string' && key.length ) {\n\t\t\t\tfromMatches[key] = fromMatches[key] || [];\n\t\t\t\tfromMatches[key].push( element );\n\t\t\t}\n\t\t} );\n\n\t\t[].slice.call( toScope.querySelectorAll( selector ) ).forEach( ( element, i ) => {\n\t\t\tconst key = serializer( element );\n\t\t\ttoMatches[key] = toMatches[key] || [];\n\t\t\ttoMatches[key].push( element );\n\n\t\t\tlet fromElement;\n\n\t\t\t// Retrieve the 'from' element\n\t\t\tif( fromMatches[key] ) {\n\t\t\t\tconst pimaryIndex = toMatches[key].length - 1;\n\t\t\t\tconst secondaryIndex = fromMatches[key].length - 1;\n\n\t\t\t\t// If there are multiple identical from elements, retrieve\n\t\t\t\t// the one at the same index as our to-element.\n\t\t\t\tif( fromMatches[key][ pimaryIndex ] ) {\n\t\t\t\t\tfromElement = fromMatches[key][ pimaryIndex ];\n\t\t\t\t\tfromMatches[key][ pimaryIndex ] = null;\n\t\t\t\t}\n\t\t\t\t// If there are no matching from-elements at the same index,\n\t\t\t\t// use the last one.\n\t\t\t\telse if( fromMatches[key][ secondaryIndex ] ) {\n\t\t\t\t\tfromElement = fromMatches[key][ secondaryIndex ];\n\t\t\t\t\tfromMatches[key][ secondaryIndex ] = null;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// If we've got a matching pair, push it to the list of pairs\n\t\t\tif( fromElement ) {\n\t\t\t\tpairs.push({\n\t\t\t\t\tfrom: fromElement,\n\t\t\t\t\tto: element,\n\t\t\t\t\toptions: animationOptions\n\t\t\t\t});\n\t\t\t}\n\t\t} );\n\n\t}\n\n\t/**\n\t * Returns a all elements within the given scope that should\n\t * be considered unmatched in an auto-animate transition. If\n\t * fading of unmatched elements is turned on, these elements\n\t * will fade when going between auto-animate slides.\n\t *\n\t * Note that parents of auto-animate targets are NOT considerd\n\t * unmatched since fading them would break the auto-animation.\n\t *\n\t * @param {HTMLElement} rootElement\n\t * @return {Array}\n\t */\n\tgetUnmatchedAutoAnimateElements( rootElement ) {\n\n\t\treturn [].slice.call( rootElement.children ).reduce( ( result, element ) => {\n\n\t\t\tconst containsAnimatedElements = element.querySelector( '[data-auto-animate-target]' );\n\n\t\t\t// The element is unmatched if\n\t\t\t// - It is not an auto-animate target\n\t\t\t// - It does not contain any auto-animate targets\n\t\t\tif( !element.hasAttribute( 'data-auto-animate-target' ) && !containsAnimatedElements ) {\n\t\t\t\tresult.push( element );\n\t\t\t}\n\n\t\t\tif( element.querySelector( '[data-auto-animate-target]' ) ) {\n\t\t\t\tresult = result.concat( this.getUnmatchedAutoAnimateElements( element ) );\n\t\t\t}\n\n\t\t\treturn result;\n\n\t\t}, [] );\n\n\t}\n\n}\n","import { extend, queryAll } from '../utils/util.js'\n\n/**\n * Handles sorting and navigation of slide fragments.\n * Fragments are elements within a slide that are\n * revealed/animated incrementally.\n */\nexport default class Fragments {\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t}\n\n\t/**\n\t * Called when the reveal.js config is updated.\n\t */\n\tconfigure( config, oldConfig ) {\n\n\t\tif( config.fragments === false ) {\n\t\t\tthis.disable();\n\t\t}\n\t\telse if( oldConfig.fragments === false ) {\n\t\t\tthis.enable();\n\t\t}\n\n\t}\n\n\t/**\n\t * If fragments are disabled in the deck, they should all be\n\t * visible rather than stepped through.\n\t */\n\tdisable() {\n\n\t\tqueryAll( this.Reveal.getSlidesElement(), '.fragment' ).forEach( element => {\n\t\t\telement.classList.add( 'visible' );\n\t\t\telement.classList.remove( 'current-fragment' );\n\t\t} );\n\n\t}\n\n\t/**\n\t * Reverse of #disable(). Only called if fragments have\n\t * previously been disabled.\n\t */\n\tenable() {\n\n\t\tqueryAll( this.Reveal.getSlidesElement(), '.fragment' ).forEach( element => {\n\t\t\telement.classList.remove( 'visible' );\n\t\t\telement.classList.remove( 'current-fragment' );\n\t\t} );\n\n\t}\n\n\t/**\n\t * Returns an object describing the available fragment\n\t * directions.\n\t *\n\t * @return {{prev: boolean, next: boolean}}\n\t */\n\tavailableRoutes() {\n\n\t\tlet currentSlide = this.Reveal.getCurrentSlide();\n\t\tif( currentSlide && this.Reveal.getConfig().fragments ) {\n\t\t\tlet fragments = currentSlide.querySelectorAll( '.fragment:not(.disabled)' );\n\t\t\tlet hiddenFragments = currentSlide.querySelectorAll( '.fragment:not(.disabled):not(.visible)' );\n\n\t\t\treturn {\n\t\t\t\tprev: fragments.length - hiddenFragments.length > 0,\n\t\t\t\tnext: !!hiddenFragments.length\n\t\t\t};\n\t\t}\n\t\telse {\n\t\t\treturn { prev: false, next: false };\n\t\t}\n\n\t}\n\n\t/**\n\t * Return a sorted fragments list, ordered by an increasing\n\t * \"data-fragment-index\" attribute.\n\t *\n\t * Fragments will be revealed in the order that they are returned by\n\t * this function, so you can use the index attributes to control the\n\t * order of fragment appearance.\n\t *\n\t * To maintain a sensible default fragment order, fragments are presumed\n\t * to be passed in document order. This function adds a \"fragment-index\"\n\t * attribute to each node if such an attribute is not already present,\n\t * and sets that attribute to an integer value which is the position of\n\t * the fragment within the fragments list.\n\t *\n\t * @param {object[]|*} fragments\n\t * @param {boolean} grouped If true the returned array will contain\n\t * nested arrays for all fragments with the same index\n\t * @return {object[]} sorted Sorted array of fragments\n\t */\n\tsort( fragments, grouped = false ) {\n\n\t\tfragments = Array.from( fragments );\n\n\t\tlet ordered = [],\n\t\t\tunordered = [],\n\t\t\tsorted = [];\n\n\t\t// Group ordered and unordered elements\n\t\tfragments.forEach( fragment => {\n\t\t\tif( fragment.hasAttribute( 'data-fragment-index' ) ) {\n\t\t\t\tlet index = parseInt( fragment.getAttribute( 'data-fragment-index' ), 10 );\n\n\t\t\t\tif( !ordered[index] ) {\n\t\t\t\t\tordered[index] = [];\n\t\t\t\t}\n\n\t\t\t\tordered[index].push( fragment );\n\t\t\t}\n\t\t\telse {\n\t\t\t\tunordered.push( [ fragment ] );\n\t\t\t}\n\t\t} );\n\n\t\t// Append fragments without explicit indices in their\n\t\t// DOM order\n\t\tordered = ordered.concat( unordered );\n\n\t\t// Manually count the index up per group to ensure there\n\t\t// are no gaps\n\t\tlet index = 0;\n\n\t\t// Push all fragments in their sorted order to an array,\n\t\t// this flattens the groups\n\t\tordered.forEach( group => {\n\t\t\tgroup.forEach( fragment => {\n\t\t\t\tsorted.push( fragment );\n\t\t\t\tfragment.setAttribute( 'data-fragment-index', index );\n\t\t\t} );\n\n\t\t\tindex ++;\n\t\t} );\n\n\t\treturn grouped === true ? ordered : sorted;\n\n\t}\n\n\t/**\n\t * Sorts and formats all of fragments in the\n\t * presentation.\n\t */\n\tsortAll() {\n\n\t\tthis.Reveal.getHorizontalSlides().forEach( horizontalSlide => {\n\n\t\t\tlet verticalSlides = queryAll( horizontalSlide, 'section' );\n\t\t\tverticalSlides.forEach( ( verticalSlide, y ) => {\n\n\t\t\t\tthis.sort( verticalSlide.querySelectorAll( '.fragment' ) );\n\n\t\t\t}, this );\n\n\t\t\tif( verticalSlides.length === 0 ) this.sort( horizontalSlide.querySelectorAll( '.fragment' ) );\n\n\t\t} );\n\n\t}\n\n\t/**\n\t * Refreshes the fragments on the current slide so that they\n\t * have the appropriate classes (.visible + .current-fragment).\n\t *\n\t * @param {number} [index] The index of the current fragment\n\t * @param {array} [fragments] Array containing all fragments\n\t * in the current slide\n\t *\n\t * @return {{shown: array, hidden: array}}\n\t */\n\tupdate( index, fragments ) {\n\n\t\tlet changedFragments = {\n\t\t\tshown: [],\n\t\t\thidden: []\n\t\t};\n\n\t\tlet currentSlide = this.Reveal.getCurrentSlide();\n\t\tif( currentSlide && this.Reveal.getConfig().fragments ) {\n\n\t\t\tfragments = fragments || this.sort( currentSlide.querySelectorAll( '.fragment' ) );\n\n\t\t\tif( fragments.length ) {\n\n\t\t\t\tlet maxIndex = 0;\n\n\t\t\t\tif( typeof index !== 'number' ) {\n\t\t\t\t\tlet currentFragment = this.sort( currentSlide.querySelectorAll( '.fragment.visible' ) ).pop();\n\t\t\t\t\tif( currentFragment ) {\n\t\t\t\t\t\tindex = parseInt( currentFragment.getAttribute( 'data-fragment-index' ) || 0, 10 );\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\tArray.from( fragments ).forEach( ( el, i ) => {\n\n\t\t\t\t\tif( el.hasAttribute( 'data-fragment-index' ) ) {\n\t\t\t\t\t\ti = parseInt( el.getAttribute( 'data-fragment-index' ), 10 );\n\t\t\t\t\t}\n\n\t\t\t\t\tmaxIndex = Math.max( maxIndex, i );\n\n\t\t\t\t\t// Visible fragments\n\t\t\t\t\tif( i <= index ) {\n\t\t\t\t\t\tlet wasVisible = el.classList.contains( 'visible' )\n\t\t\t\t\t\tel.classList.add( 'visible' );\n\t\t\t\t\t\tel.classList.remove( 'current-fragment' );\n\n\t\t\t\t\t\tif( i === index ) {\n\t\t\t\t\t\t\t// Announce the fragments one by one to the Screen Reader\n\t\t\t\t\t\t\tthis.Reveal.announceStatus( this.Reveal.getStatusText( el ) );\n\n\t\t\t\t\t\t\tel.classList.add( 'current-fragment' );\n\t\t\t\t\t\t\tthis.Reveal.slideContent.startEmbeddedContent( el );\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tif( !wasVisible ) {\n\t\t\t\t\t\t\tchangedFragments.shown.push( el )\n\t\t\t\t\t\t\tthis.Reveal.dispatchEvent({\n\t\t\t\t\t\t\t\ttarget: el,\n\t\t\t\t\t\t\t\ttype: 'visible',\n\t\t\t\t\t\t\t\tbubbles: false\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\t// Hidden fragments\n\t\t\t\t\telse {\n\t\t\t\t\t\tlet wasVisible = el.classList.contains( 'visible' )\n\t\t\t\t\t\tel.classList.remove( 'visible' );\n\t\t\t\t\t\tel.classList.remove( 'current-fragment' );\n\n\t\t\t\t\t\tif( wasVisible ) {\n\t\t\t\t\t\t\tthis.Reveal.slideContent.stopEmbeddedContent( el );\n\t\t\t\t\t\t\tchangedFragments.hidden.push( el );\n\t\t\t\t\t\t\tthis.Reveal.dispatchEvent({\n\t\t\t\t\t\t\t\ttarget: el,\n\t\t\t\t\t\t\t\ttype: 'hidden',\n\t\t\t\t\t\t\t\tbubbles: false\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t} );\n\n\t\t\t\t// Write the current fragment index to the slide .\n\t\t\t\t// This can be used by end users to apply styles based on\n\t\t\t\t// the current fragment index.\n\t\t\t\tindex = typeof index === 'number' ? index : -1;\n\t\t\t\tindex = Math.max( Math.min( index, maxIndex ), -1 );\n\t\t\t\tcurrentSlide.setAttribute( 'data-fragment', index );\n\n\t\t\t}\n\n\t\t}\n\n\t\treturn changedFragments;\n\n\t}\n\n\t/**\n\t * Formats the fragments on the given slide so that they have\n\t * valid indices. Call this if fragments are changed in the DOM\n\t * after reveal.js has already initialized.\n\t *\n\t * @param {HTMLElement} slide\n\t * @return {Array} a list of the HTML fragments that were synced\n\t */\n\tsync( slide = this.Reveal.getCurrentSlide() ) {\n\n\t\treturn this.sort( slide.querySelectorAll( '.fragment' ) );\n\n\t}\n\n\t/**\n\t * Navigate to the specified slide fragment.\n\t *\n\t * @param {?number} index The index of the fragment that\n\t * should be shown, -1 means all are invisible\n\t * @param {number} offset Integer offset to apply to the\n\t * fragment index\n\t *\n\t * @return {boolean} true if a change was made in any\n\t * fragments visibility as part of this call\n\t */\n\tgoto( index, offset = 0 ) {\n\n\t\tlet currentSlide = this.Reveal.getCurrentSlide();\n\t\tif( currentSlide && this.Reveal.getConfig().fragments ) {\n\n\t\t\tlet fragments = this.sort( currentSlide.querySelectorAll( '.fragment:not(.disabled)' ) );\n\t\t\tif( fragments.length ) {\n\n\t\t\t\t// If no index is specified, find the current\n\t\t\t\tif( typeof index !== 'number' ) {\n\t\t\t\t\tlet lastVisibleFragment = this.sort( currentSlide.querySelectorAll( '.fragment:not(.disabled).visible' ) ).pop();\n\n\t\t\t\t\tif( lastVisibleFragment ) {\n\t\t\t\t\t\tindex = parseInt( lastVisibleFragment.getAttribute( 'data-fragment-index' ) || 0, 10 );\n\t\t\t\t\t}\n\t\t\t\t\telse {\n\t\t\t\t\t\tindex = -1;\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\t// Apply the offset if there is one\n\t\t\t\tindex += offset;\n\n\t\t\t\tlet changedFragments = this.update( index, fragments );\n\n\t\t\t\tif( changedFragments.hidden.length ) {\n\t\t\t\t\tthis.Reveal.dispatchEvent({\n\t\t\t\t\t\ttype: 'fragmenthidden',\n\t\t\t\t\t\tdata: {\n\t\t\t\t\t\t\tfragment: changedFragments.hidden[0],\n\t\t\t\t\t\t\tfragments: changedFragments.hidden\n\t\t\t\t\t\t}\n\t\t\t\t\t});\n\t\t\t\t}\n\n\t\t\t\tif( changedFragments.shown.length ) {\n\t\t\t\t\tthis.Reveal.dispatchEvent({\n\t\t\t\t\t\ttype: 'fragmentshown',\n\t\t\t\t\t\tdata: {\n\t\t\t\t\t\t\tfragment: changedFragments.shown[0],\n\t\t\t\t\t\t\tfragments: changedFragments.shown\n\t\t\t\t\t\t}\n\t\t\t\t\t});\n\t\t\t\t}\n\n\t\t\t\tthis.Reveal.controls.update();\n\t\t\t\tthis.Reveal.progress.update();\n\n\t\t\t\tif( this.Reveal.getConfig().fragmentInURL ) {\n\t\t\t\t\tthis.Reveal.location.writeURL();\n\t\t\t\t}\n\n\t\t\t\treturn !!( changedFragments.shown.length || changedFragments.hidden.length );\n\n\t\t\t}\n\n\t\t}\n\n\t\treturn false;\n\n\t}\n\n\t/**\n\t * Navigate to the next slide fragment.\n\t *\n\t * @return {boolean} true if there was a next fragment,\n\t * false otherwise\n\t */\n\tnext() {\n\n\t\treturn this.goto( null, 1 );\n\n\t}\n\n\t/**\n\t * Navigate to the previous slide fragment.\n\t *\n\t * @return {boolean} true if there was a previous fragment,\n\t * false otherwise\n\t */\n\tprev() {\n\n\t\treturn this.goto( null, -1 );\n\n\t}\n\n}","import { SLIDES_SELECTOR } from '../utils/constants.js'\nimport { extend, queryAll, transformElement } from '../utils/util.js'\n\n/**\n * Handles all logic related to the overview mode\n * (birds-eye view of all slides).\n */\nexport default class Overview {\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t\tthis.active = false;\n\n\t\tthis.onSlideClicked = this.onSlideClicked.bind( this );\n\n\t}\n\n\t/**\n\t * Displays the overview of slides (quick nav) by scaling\n\t * down and arranging all slide elements.\n\t */\n\tactivate() {\n\n\t\t// Only proceed if enabled in config\n\t\tif( this.Reveal.getConfig().overview && !this.isActive() ) {\n\n\t\t\tthis.active = true;\n\n\t\t\tthis.Reveal.getRevealElement().classList.add( 'overview' );\n\n\t\t\t// Don't auto-slide while in overview mode\n\t\t\tthis.Reveal.cancelAutoSlide();\n\n\t\t\t// Move the backgrounds element into the slide container to\n\t\t\t// that the same scaling is applied\n\t\t\tthis.Reveal.getSlidesElement().appendChild( this.Reveal.getBackgroundsElement() );\n\n\t\t\t// Clicking on an overview slide navigates to it\n\t\t\tqueryAll( this.Reveal.getRevealElement(), SLIDES_SELECTOR ).forEach( slide => {\n\t\t\t\tif( !slide.classList.contains( 'stack' ) ) {\n\t\t\t\t\tslide.addEventListener( 'click', this.onSlideClicked, true );\n\t\t\t\t}\n\t\t\t} );\n\n\t\t\t// Calculate slide sizes\n\t\t\tconst margin = 70;\n\t\t\tconst slideSize = this.Reveal.getComputedSlideSize();\n\t\t\tthis.overviewSlideWidth = slideSize.width + margin;\n\t\t\tthis.overviewSlideHeight = slideSize.height + margin;\n\n\t\t\t// Reverse in RTL mode\n\t\t\tif( this.Reveal.getConfig().rtl ) {\n\t\t\t\tthis.overviewSlideWidth = -this.overviewSlideWidth;\n\t\t\t}\n\n\t\t\tthis.Reveal.updateSlidesVisibility();\n\n\t\t\tthis.layout();\n\t\t\tthis.update();\n\n\t\t\tthis.Reveal.layout();\n\n\t\t\tconst indices = this.Reveal.getIndices();\n\n\t\t\t// Notify observers of the overview showing\n\t\t\tthis.Reveal.dispatchEvent({\n\t\t\t\ttype: 'overviewshown',\n\t\t\t\tdata: {\n\t\t\t\t\t'indexh': indices.h,\n\t\t\t\t\t'indexv': indices.v,\n\t\t\t\t\t'currentSlide': this.Reveal.getCurrentSlide()\n\t\t\t\t}\n\t\t\t});\n\n\t\t}\n\n\t}\n\n\t/**\n\t * Uses CSS transforms to position all slides in a grid for\n\t * display inside of the overview mode.\n\t */\n\tlayout() {\n\n\t\t// Layout slides\n\t\tthis.Reveal.getHorizontalSlides().forEach( ( hslide, h ) => {\n\t\t\thslide.setAttribute( 'data-index-h', h );\n\t\t\ttransformElement( hslide, 'translate3d(' + ( h * this.overviewSlideWidth ) + 'px, 0, 0)' );\n\n\t\t\tif( hslide.classList.contains( 'stack' ) ) {\n\n\t\t\t\tqueryAll( hslide, 'section' ).forEach( ( vslide, v ) => {\n\t\t\t\t\tvslide.setAttribute( 'data-index-h', h );\n\t\t\t\t\tvslide.setAttribute( 'data-index-v', v );\n\n\t\t\t\t\ttransformElement( vslide, 'translate3d(0, ' + ( v * this.overviewSlideHeight ) + 'px, 0)' );\n\t\t\t\t} );\n\n\t\t\t}\n\t\t} );\n\n\t\t// Layout slide backgrounds\n\t\tArray.from( this.Reveal.getBackgroundsElement().childNodes ).forEach( ( hbackground, h ) => {\n\t\t\ttransformElement( hbackground, 'translate3d(' + ( h * this.overviewSlideWidth ) + 'px, 0, 0)' );\n\n\t\t\tqueryAll( hbackground, '.slide-background' ).forEach( ( vbackground, v ) => {\n\t\t\t\ttransformElement( vbackground, 'translate3d(0, ' + ( v * this.overviewSlideHeight ) + 'px, 0)' );\n\t\t\t} );\n\t\t} );\n\n\t}\n\n\t/**\n\t * Moves the overview viewport to the current slides.\n\t * Called each time the current slide changes.\n\t */\n\tupdate() {\n\n\t\tconst vmin = Math.min( window.innerWidth, window.innerHeight );\n\t\tconst scale = Math.max( vmin / 5, 150 ) / vmin;\n\t\tconst indices = this.Reveal.getIndices();\n\n\t\tthis.Reveal.transformSlides( {\n\t\t\toverview: [\n\t\t\t\t'scale('+ scale +')',\n\t\t\t\t'translateX('+ ( -indices.h * this.overviewSlideWidth ) +'px)',\n\t\t\t\t'translateY('+ ( -indices.v * this.overviewSlideHeight ) +'px)'\n\t\t\t].join( ' ' )\n\t\t} );\n\n\t}\n\n\t/**\n\t * Exits the slide overview and enters the currently\n\t * active slide.\n\t */\n\tdeactivate() {\n\n\t\t// Only proceed if enabled in config\n\t\tif( this.Reveal.getConfig().overview ) {\n\n\t\t\tthis.active = false;\n\n\t\t\tthis.Reveal.getRevealElement().classList.remove( 'overview' );\n\n\t\t\t// Temporarily add a class so that transitions can do different things\n\t\t\t// depending on whether they are exiting/entering overview, or just\n\t\t\t// moving from slide to slide\n\t\t\tthis.Reveal.getRevealElement().classList.add( 'overview-deactivating' );\n\n\t\t\tsetTimeout( () => {\n\t\t\t\tthis.Reveal.getRevealElement().classList.remove( 'overview-deactivating' );\n\t\t\t}, 1 );\n\n\t\t\t// Move the background element back out\n\t\t\tthis.Reveal.getRevealElement().appendChild( this.Reveal.getBackgroundsElement() );\n\n\t\t\t// Clean up changes made to slides\n\t\t\tqueryAll( this.Reveal.getRevealElement(), SLIDES_SELECTOR ).forEach( slide => {\n\t\t\t\ttransformElement( slide, '' );\n\n\t\t\t\tslide.removeEventListener( 'click', this.onSlideClicked, true );\n\t\t\t} );\n\n\t\t\t// Clean up changes made to backgrounds\n\t\t\tqueryAll( this.Reveal.getBackgroundsElement(), '.slide-background' ).forEach( background => {\n\t\t\t\ttransformElement( background, '' );\n\t\t\t} );\n\n\t\t\tthis.Reveal.transformSlides( { overview: '' } );\n\n\t\t\tconst indices = this.Reveal.getIndices();\n\n\t\t\tthis.Reveal.slide( indices.h, indices.v );\n\t\t\tthis.Reveal.layout();\n\t\t\tthis.Reveal.cueAutoSlide();\n\n\t\t\t// Notify observers of the overview hiding\n\t\t\tthis.Reveal.dispatchEvent({\n\t\t\t\ttype: 'overviewhidden',\n\t\t\t\tdata: {\n\t\t\t\t\t'indexh': indices.h,\n\t\t\t\t\t'indexv': indices.v,\n\t\t\t\t\t'currentSlide': this.Reveal.getCurrentSlide()\n\t\t\t\t}\n\t\t\t});\n\n\t\t}\n\t}\n\n\t/**\n\t * Toggles the slide overview mode on and off.\n\t *\n\t * @param {Boolean} [override] Flag which overrides the\n\t * toggle logic and forcibly sets the desired state. True means\n\t * overview is open, false means it's closed.\n\t */\n\ttoggle( override ) {\n\n\t\tif( typeof override === 'boolean' ) {\n\t\t\toverride ? this.activate() : this.deactivate();\n\t\t}\n\t\telse {\n\t\t\tthis.isActive() ? this.deactivate() : this.activate();\n\t\t}\n\n\t}\n\n\t/**\n\t * Checks if the overview is currently active.\n\t *\n\t * @return {Boolean} true if the overview is active,\n\t * false otherwise\n\t */\n\tisActive() {\n\n\t\treturn this.active;\n\n\t}\n\n\t/**\n\t * Invoked when a slide is and we're in the overview.\n\t *\n\t * @param {object} event\n\t */\n\tonSlideClicked( event ) {\n\n\t\tif( this.isActive() ) {\n\t\t\tevent.preventDefault();\n\n\t\t\tlet element = event.target;\n\n\t\t\twhile( element && !element.nodeName.match( /section/gi ) ) {\n\t\t\t\telement = element.parentNode;\n\t\t\t}\n\n\t\t\tif( element && !element.classList.contains( 'disabled' ) ) {\n\n\t\t\t\tthis.deactivate();\n\n\t\t\t\tif( element.nodeName.match( /section/gi ) ) {\n\t\t\t\t\tlet h = parseInt( element.getAttribute( 'data-index-h' ), 10 ),\n\t\t\t\t\t\tv = parseInt( element.getAttribute( 'data-index-v' ), 10 );\n\n\t\t\t\t\tthis.Reveal.slide( h, v );\n\t\t\t\t}\n\n\t\t\t}\n\t\t}\n\n\t}\n\n}","import { enterFullscreen } from '../utils/util.js'\n\n/**\n * Handles all reveal.js keyboard interactions.\n */\nexport default class Keyboard {\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t\t// A key:value map of keyboard keys and descriptions of\n\t\t// the actions they trigger\n\t\tthis.shortcuts = {};\n\n\t\t// Holds custom key code mappings\n\t\tthis.bindings = {};\n\n\t\tthis.onDocumentKeyDown = this.onDocumentKeyDown.bind( this );\n\t\tthis.onDocumentKeyPress = this.onDocumentKeyPress.bind( this );\n\n\t}\n\n\t/**\n\t * Called when the reveal.js config is updated.\n\t */\n\tconfigure( config, oldConfig ) {\n\n\t\tif( config.navigationMode === 'linear' ) {\n\t\t\tthis.shortcuts['→ , ↓ , SPACE , N , L , J'] = 'Next slide';\n\t\t\tthis.shortcuts['← , ↑ , P , H , K'] = 'Previous slide';\n\t\t}\n\t\telse {\n\t\t\tthis.shortcuts['N , SPACE'] = 'Next slide';\n\t\t\tthis.shortcuts['P , Shift SPACE'] = 'Previous slide';\n\t\t\tthis.shortcuts['← , H'] = 'Navigate left';\n\t\t\tthis.shortcuts['→ , L'] = 'Navigate right';\n\t\t\tthis.shortcuts['↑ , K'] = 'Navigate up';\n\t\t\tthis.shortcuts['↓ , J'] = 'Navigate down';\n\t\t}\n\n\t\tthis.shortcuts['Alt + ←/↑/→/↓'] = 'Navigate without fragments';\n\t\tthis.shortcuts['Shift + ←/↑/→/↓'] = 'Jump to first/last slide';\n\t\tthis.shortcuts['B , .'] = 'Pause';\n\t\tthis.shortcuts['F'] = 'Fullscreen';\n\t\tthis.shortcuts['ESC, O'] = 'Slide overview';\n\n\t}\n\n\t/**\n\t * Starts listening for keyboard events.\n\t */\n\tbind() {\n\n\t\tdocument.addEventListener( 'keydown', this.onDocumentKeyDown, false );\n\t\tdocument.addEventListener( 'keypress', this.onDocumentKeyPress, false );\n\n\t}\n\n\t/**\n\t * Stops listening for keyboard events.\n\t */\n\tunbind() {\n\n\t\tdocument.removeEventListener( 'keydown', this.onDocumentKeyDown, false );\n\t\tdocument.removeEventListener( 'keypress', this.onDocumentKeyPress, false );\n\n\t}\n\n\t/**\n\t * Add a custom key binding with optional description to\n\t * be added to the help screen.\n\t */\n\taddKeyBinding( binding, callback ) {\n\n\t\tif( typeof binding === 'object' && binding.keyCode ) {\n\t\t\tthis.bindings[binding.keyCode] = {\n\t\t\t\tcallback: callback,\n\t\t\t\tkey: binding.key,\n\t\t\t\tdescription: binding.description\n\t\t\t};\n\t\t}\n\t\telse {\n\t\t\tthis.bindings[binding] = {\n\t\t\t\tcallback: callback,\n\t\t\t\tkey: null,\n\t\t\t\tdescription: null\n\t\t\t};\n\t\t}\n\n\t}\n\n\t/**\n\t * Removes the specified custom key binding.\n\t */\n\tremoveKeyBinding( keyCode ) {\n\n\t\tdelete this.bindings[keyCode];\n\n\t}\n\n\t/**\n\t * Programmatically triggers a keyboard event\n\t *\n\t * @param {int} keyCode\n\t */\n\ttriggerKey( keyCode ) {\n\n\t\tthis.onDocumentKeyDown( { keyCode } );\n\n\t}\n\n\t/**\n\t * Registers a new shortcut to include in the help overlay\n\t *\n\t * @param {String} key\n\t * @param {String} value\n\t */\n\tregisterKeyboardShortcut( key, value ) {\n\n\t\tthis.shortcuts[key] = value;\n\n\t}\n\n\tgetShortcuts() {\n\n\t\treturn this.shortcuts;\n\n\t}\n\n\tgetBindings() {\n\n\t\treturn this.bindings;\n\n\t}\n\n\t/**\n\t * Handler for the document level 'keypress' event.\n\t *\n\t * @param {object} event\n\t */\n\tonDocumentKeyPress( event ) {\n\n\t\t// Check if the pressed key is question mark\n\t\tif( event.shiftKey && event.charCode === 63 ) {\n\t\t\tthis.Reveal.toggleHelp();\n\t\t}\n\n\t}\n\n\t/**\n\t * Handler for the document level 'keydown' event.\n\t *\n\t * @param {object} event\n\t */\n\tonDocumentKeyDown( event ) {\n\n\t\tlet config = this.Reveal.getConfig();\n\n\t\t// If there's a condition specified and it returns false,\n\t\t// ignore this event\n\t\tif( typeof config.keyboardCondition === 'function' && config.keyboardCondition(event) === false ) {\n\t\t\treturn true;\n\t\t}\n\n\t\t// If keyboardCondition is set, only capture keyboard events\n\t\t// for embedded decks when they are focused\n\t\tif( config.keyboardCondition === 'focused' && !this.Reveal.isFocused() ) {\n\t\t\treturn true;\n\t\t}\n\n\t\t// Shorthand\n\t\tlet keyCode = event.keyCode;\n\n\t\t// Remember if auto-sliding was paused so we can toggle it\n\t\tlet autoSlideWasPaused = !this.Reveal.isAutoSliding();\n\n\t\tthis.Reveal.onUserInput( event );\n\n\t\t// Is there a focused element that could be using the keyboard?\n\t\tlet activeElementIsCE = document.activeElement && document.activeElement.isContentEditable === true;\n\t\tlet activeElementIsInput = document.activeElement && document.activeElement.tagName && /input|textarea/i.test( document.activeElement.tagName );\n\t\tlet activeElementIsNotes = document.activeElement && document.activeElement.className && /speaker-notes/i.test( document.activeElement.className);\n\n\t\t// Whitelist certain modifiers for slide navigation shortcuts\n\t\tlet isNavigationKey = [32, 37, 38, 39, 40, 78, 80].indexOf( event.keyCode ) !== -1;\n\n\t\t// Prevent all other events when a modifier is pressed\n\t\tlet unusedModifier = \t!( isNavigationKey && event.shiftKey || event.altKey ) &&\n\t\t\t\t\t\t\t\t( event.shiftKey || event.altKey || event.ctrlKey || event.metaKey );\n\n\t\t// Disregard the event if there's a focused element or a\n\t\t// keyboard modifier key is present\n\t\tif( activeElementIsCE || activeElementIsInput || activeElementIsNotes || unusedModifier ) return;\n\n\t\t// While paused only allow resume keyboard events; 'b', 'v', '.'\n\t\tlet resumeKeyCodes = [66,86,190,191];\n\t\tlet key;\n\n\t\t// Custom key bindings for togglePause should be able to resume\n\t\tif( typeof config.keyboard === 'object' ) {\n\t\t\tfor( key in config.keyboard ) {\n\t\t\t\tif( config.keyboard[key] === 'togglePause' ) {\n\t\t\t\t\tresumeKeyCodes.push( parseInt( key, 10 ) );\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tif( this.Reveal.isPaused() && resumeKeyCodes.indexOf( keyCode ) === -1 ) {\n\t\t\treturn false;\n\t\t}\n\n\t\t// Use linear navigation if we're configured to OR if\n\t\t// the presentation is one-dimensional\n\t\tlet useLinearMode = config.navigationMode === 'linear' || !this.Reveal.hasHorizontalSlides() || !this.Reveal.hasVerticalSlides();\n\n\t\tlet triggered = false;\n\n\t\t// 1. User defined key bindings\n\t\tif( typeof config.keyboard === 'object' ) {\n\n\t\t\tfor( key in config.keyboard ) {\n\n\t\t\t\t// Check if this binding matches the pressed key\n\t\t\t\tif( parseInt( key, 10 ) === keyCode ) {\n\n\t\t\t\t\tlet value = config.keyboard[ key ];\n\n\t\t\t\t\t// Callback function\n\t\t\t\t\tif( typeof value === 'function' ) {\n\t\t\t\t\t\tvalue.apply( null, [ event ] );\n\t\t\t\t\t}\n\t\t\t\t\t// String shortcuts to reveal.js API\n\t\t\t\t\telse if( typeof value === 'string' && typeof this.Reveal[ value ] === 'function' ) {\n\t\t\t\t\t\tthis.Reveal[ value ].call();\n\t\t\t\t\t}\n\n\t\t\t\t\ttriggered = true;\n\n\t\t\t\t}\n\n\t\t\t}\n\n\t\t}\n\n\t\t// 2. Registered custom key bindings\n\t\tif( triggered === false ) {\n\n\t\t\tfor( key in this.bindings ) {\n\n\t\t\t\t// Check if this binding matches the pressed key\n\t\t\t\tif( parseInt( key, 10 ) === keyCode ) {\n\n\t\t\t\t\tlet action = this.bindings[ key ].callback;\n\n\t\t\t\t\t// Callback function\n\t\t\t\t\tif( typeof action === 'function' ) {\n\t\t\t\t\t\taction.apply( null, [ event ] );\n\t\t\t\t\t}\n\t\t\t\t\t// String shortcuts to reveal.js API\n\t\t\t\t\telse if( typeof action === 'string' && typeof this.Reveal[ action ] === 'function' ) {\n\t\t\t\t\t\tthis.Reveal[ action ].call();\n\t\t\t\t\t}\n\n\t\t\t\t\ttriggered = true;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// 3. System defined key bindings\n\t\tif( triggered === false ) {\n\n\t\t\t// Assume true and try to prove false\n\t\t\ttriggered = true;\n\n\t\t\t// P, PAGE UP\n\t\t\tif( keyCode === 80 || keyCode === 33 ) {\n\t\t\t\tthis.Reveal.prev({skipFragments: event.altKey});\n\t\t\t}\n\t\t\t// N, PAGE DOWN\n\t\t\telse if( keyCode === 78 || keyCode === 34 ) {\n\t\t\t\tthis.Reveal.next({skipFragments: event.altKey});\n\t\t\t}\n\t\t\t// H, LEFT\n\t\t\telse if( keyCode === 72 || keyCode === 37 ) {\n\t\t\t\tif( event.shiftKey ) {\n\t\t\t\t\tthis.Reveal.slide( 0 );\n\t\t\t\t}\n\t\t\t\telse if( !this.Reveal.overview.isActive() && useLinearMode ) {\n\t\t\t\t\tthis.Reveal.prev({skipFragments: event.altKey});\n\t\t\t\t}\n\t\t\t\telse {\n\t\t\t\t\tthis.Reveal.left({skipFragments: event.altKey});\n\t\t\t\t}\n\t\t\t}\n\t\t\t// L, RIGHT\n\t\t\telse if( keyCode === 76 || keyCode === 39 ) {\n\t\t\t\tif( event.shiftKey ) {\n\t\t\t\t\tthis.Reveal.slide( this.Reveal.getHorizontalSlides().length - 1 );\n\t\t\t\t}\n\t\t\t\telse if( !this.Reveal.overview.isActive() && useLinearMode ) {\n\t\t\t\t\tthis.Reveal.next({skipFragments: event.altKey});\n\t\t\t\t}\n\t\t\t\telse {\n\t\t\t\t\tthis.Reveal.right({skipFragments: event.altKey});\n\t\t\t\t}\n\t\t\t}\n\t\t\t// K, UP\n\t\t\telse if( keyCode === 75 || keyCode === 38 ) {\n\t\t\t\tif( event.shiftKey ) {\n\t\t\t\t\tthis.Reveal.slide( undefined, 0 );\n\t\t\t\t}\n\t\t\t\telse if( !this.Reveal.overview.isActive() && useLinearMode ) {\n\t\t\t\t\tthis.Reveal.prev({skipFragments: event.altKey});\n\t\t\t\t}\n\t\t\t\telse {\n\t\t\t\t\tthis.Reveal.up({skipFragments: event.altKey});\n\t\t\t\t}\n\t\t\t}\n\t\t\t// J, DOWN\n\t\t\telse if( keyCode === 74 || keyCode === 40 ) {\n\t\t\t\tif( event.shiftKey ) {\n\t\t\t\t\tthis.Reveal.slide( undefined, Number.MAX_VALUE );\n\t\t\t\t}\n\t\t\t\telse if( !this.Reveal.overview.isActive() && useLinearMode ) {\n\t\t\t\t\tthis.Reveal.next({skipFragments: event.altKey});\n\t\t\t\t}\n\t\t\t\telse {\n\t\t\t\t\tthis.Reveal.down({skipFragments: event.altKey});\n\t\t\t\t}\n\t\t\t}\n\t\t\t// HOME\n\t\t\telse if( keyCode === 36 ) {\n\t\t\t\tthis.Reveal.slide( 0 );\n\t\t\t}\n\t\t\t// END\n\t\t\telse if( keyCode === 35 ) {\n\t\t\t\tthis.Reveal.slide( this.Reveal.getHorizontalSlides().length - 1 );\n\t\t\t}\n\t\t\t// SPACE\n\t\t\telse if( keyCode === 32 ) {\n\t\t\t\tif( this.Reveal.overview.isActive() ) {\n\t\t\t\t\tthis.Reveal.overview.deactivate();\n\t\t\t\t}\n\t\t\t\tif( event.shiftKey ) {\n\t\t\t\t\tthis.Reveal.prev({skipFragments: event.altKey});\n\t\t\t\t}\n\t\t\t\telse {\n\t\t\t\t\tthis.Reveal.next({skipFragments: event.altKey});\n\t\t\t\t}\n\t\t\t}\n\t\t\t// TWO-SPOT, SEMICOLON, B, V, PERIOD, LOGITECH PRESENTER TOOLS \"BLACK SCREEN\" BUTTON\n\t\t\telse if( keyCode === 58 || keyCode === 59 || keyCode === 66 || keyCode === 86 || keyCode === 190 || keyCode === 191 ) {\n\t\t\t\tthis.Reveal.togglePause();\n\t\t\t}\n\t\t\t// F\n\t\t\telse if( keyCode === 70 ) {\n\t\t\t\tenterFullscreen( config.embedded ? this.Reveal.getViewportElement() : document.documentElement );\n\t\t\t}\n\t\t\t// A\n\t\t\telse if( keyCode === 65 ) {\n\t\t\t\tif ( config.autoSlideStoppable ) {\n\t\t\t\t\tthis.Reveal.toggleAutoSlide( autoSlideWasPaused );\n\t\t\t\t}\n\t\t\t}\n\t\t\telse {\n\t\t\t\ttriggered = false;\n\t\t\t}\n\n\t\t}\n\n\t\t// If the input resulted in a triggered action we should prevent\n\t\t// the browsers default behavior\n\t\tif( triggered ) {\n\t\t\tevent.preventDefault && event.preventDefault();\n\t\t}\n\t\t// ESC or O key\n\t\telse if( keyCode === 27 || keyCode === 79 ) {\n\t\t\tif( this.Reveal.closeOverlay() === false ) {\n\t\t\t\tthis.Reveal.overview.toggle();\n\t\t\t}\n\n\t\t\tevent.preventDefault && event.preventDefault();\n\t\t}\n\n\t\t// If auto-sliding is enabled we need to cue up\n\t\t// another timeout\n\t\tthis.Reveal.cueAutoSlide();\n\n\t}\n\n}","/**\n * Reads and writes the URL based on reveal.js' current state.\n */\nexport default class Location {\n\n\t// The minimum number of milliseconds that must pass between\n\t// calls to history.replaceState\n\tMAX_REPLACE_STATE_FREQUENCY = 1000\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t\t// Delays updates to the URL due to a Chrome thumbnailer bug\n\t\tthis.writeURLTimeout = 0;\n\n\t\tthis.replaceStateTimestamp = 0;\n\n\t\tthis.onWindowHashChange = this.onWindowHashChange.bind( this );\n\n\t}\n\n\tbind() {\n\n\t\twindow.addEventListener( 'hashchange', this.onWindowHashChange, false );\n\n\t}\n\n\tunbind() {\n\n\t\twindow.removeEventListener( 'hashchange', this.onWindowHashChange, false );\n\n\t}\n\n\t/**\n\t * Returns the slide indices for the given hash link.\n\t *\n\t * @param {string} [hash] the hash string that we want to\n\t * find the indices for\n\t *\n\t * @returns slide indices or null\n\t */\n\tgetIndicesFromHash( hash=window.location.hash ) {\n\n\t\t// Attempt to parse the hash as either an index or name\n\t\tlet name = hash.replace( /^#\\/?/, '' );\n\t\tlet bits = name.split( '/' );\n\n\t\t// If the first bit is not fully numeric and there is a name we\n\t\t// can assume that this is a named link\n\t\tif( !/^[0-9]*$/.test( bits[0] ) && name.length ) {\n\t\t\tlet element;\n\n\t\t\tlet f;\n\n\t\t\t// Parse named links with fragments (#/named-link/2)\n\t\t\tif( /\\/[-\\d]+$/g.test( name ) ) {\n\t\t\t\tf = parseInt( name.split( '/' ).pop(), 10 );\n\t\t\t\tf = isNaN(f) ? undefined : f;\n\t\t\t\tname = name.split( '/' ).shift();\n\t\t\t}\n\n\t\t\t// Ensure the named link is a valid HTML ID attribute\n\t\t\ttry {\n\t\t\t\telement = document.getElementById( decodeURIComponent( name ) );\n\t\t\t}\n\t\t\tcatch ( error ) { }\n\n\t\t\tif( element ) {\n\t\t\t\treturn { ...this.Reveal.getIndices( element ), f };\n\t\t\t}\n\t\t}\n\t\telse {\n\t\t\tconst config = this.Reveal.getConfig();\n\t\t\tlet hashIndexBase = config.hashOneBasedIndex ? 1 : 0;\n\n\t\t\t// Read the index components of the hash\n\t\t\tlet h = ( parseInt( bits[0], 10 ) - hashIndexBase ) || 0,\n\t\t\t\tv = ( parseInt( bits[1], 10 ) - hashIndexBase ) || 0,\n\t\t\t\tf;\n\n\t\t\tif( config.fragmentInURL ) {\n\t\t\t\tf = parseInt( bits[2], 10 );\n\t\t\t\tif( isNaN( f ) ) {\n\t\t\t\t\tf = undefined;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\treturn { h, v, f };\n\t\t}\n\n\t\t// The hash couldn't be parsed or no matching named link was found\n\t\treturn null\n\n\t}\n\n\t/**\n\t * Reads the current URL (hash) and navigates accordingly.\n\t */\n\treadURL() {\n\n\t\tconst currentIndices = this.Reveal.getIndices();\n\t\tconst newIndices = this.getIndicesFromHash();\n\n\t\tif( newIndices ) {\n\t\t\tif( ( newIndices.h !== currentIndices.h || newIndices.v !== currentIndices.v || newIndices.f !== undefined ) ) {\n\t\t\t\t\tthis.Reveal.slide( newIndices.h, newIndices.v, newIndices.f );\n\t\t\t}\n\t\t}\n\t\t// If no new indices are available, we're trying to navigate to\n\t\t// a slide hash that does not exist\n\t\telse {\n\t\t\tthis.Reveal.slide( currentIndices.h || 0, currentIndices.v || 0 );\n\t\t}\n\n\t}\n\n\t/**\n\t * Updates the page URL (hash) to reflect the current\n\t * state.\n\t *\n\t * @param {number} delay The time in ms to wait before\n\t * writing the hash\n\t */\n\twriteURL( delay ) {\n\n\t\tlet config = this.Reveal.getConfig();\n\t\tlet currentSlide = this.Reveal.getCurrentSlide();\n\n\t\t// Make sure there's never more than one timeout running\n\t\tclearTimeout( this.writeURLTimeout );\n\n\t\t// If a delay is specified, timeout this call\n\t\tif( typeof delay === 'number' ) {\n\t\t\tthis.writeURLTimeout = setTimeout( this.writeURL, delay );\n\t\t}\n\t\telse if( currentSlide ) {\n\n\t\t\tlet hash = this.getHash();\n\n\t\t\t// If we're configured to push to history OR the history\n\t\t\t// API is not avaialble.\n\t\t\tif( config.history ) {\n\t\t\t\twindow.location.hash = hash;\n\t\t\t}\n\t\t\t// If we're configured to reflect the current slide in the\n\t\t\t// URL without pushing to history.\n\t\t\telse if( config.hash ) {\n\t\t\t\t// If the hash is empty, don't add it to the URL\n\t\t\t\tif( hash === '/' ) {\n\t\t\t\t\tthis.debouncedReplaceState( window.location.pathname + window.location.search );\n\t\t\t\t}\n\t\t\t\telse {\n\t\t\t\t\tthis.debouncedReplaceState( '#' + hash );\n\t\t\t\t}\n\t\t\t}\n\t\t\t// UPDATE: The below nuking of all hash changes breaks\n\t\t\t// anchors on pages where reveal.js is running. Removed\n\t\t\t// in 4.0. Why was it here in the first place? ¯\\_(ツ)_/¯\n\t\t\t//\n\t\t\t// If history and hash are both disabled, a hash may still\n\t\t\t// be added to the URL by clicking on a href with a hash\n\t\t\t// target. Counter this by always removing the hash.\n\t\t\t// else {\n\t\t\t// \twindow.history.replaceState( null, null, window.location.pathname + window.location.search );\n\t\t\t// }\n\n\t\t}\n\n\t}\n\n\treplaceState( url ) {\n\n\t\twindow.history.replaceState( null, null, url );\n\t\tthis.replaceStateTimestamp = Date.now();\n\n\t}\n\n\tdebouncedReplaceState( url ) {\n\n\t\tclearTimeout( this.replaceStateTimeout );\n\n\t\tif( Date.now() - this.replaceStateTimestamp > this.MAX_REPLACE_STATE_FREQUENCY ) {\n\t\t\tthis.replaceState( url );\n\t\t}\n\t\telse {\n\t\t\tthis.replaceStateTimeout = setTimeout( () => this.replaceState( url ), this.MAX_REPLACE_STATE_FREQUENCY );\n\t\t}\n\n\t}\n\n\t/**\n\t * Return a hash URL that will resolve to the given slide location.\n\t *\n\t * @param {HTMLElement} [slide=currentSlide] The slide to link to\n\t */\n\tgetHash( slide ) {\n\n\t\tlet url = '/';\n\n\t\t// Attempt to create a named link based on the slide's ID\n\t\tlet s = slide || this.Reveal.getCurrentSlide();\n\t\tlet id = s ? s.getAttribute( 'id' ) : null;\n\t\tif( id ) {\n\t\t\tid = encodeURIComponent( id );\n\t\t}\n\n\t\tlet index = this.Reveal.getIndices( slide );\n\t\tif( !this.Reveal.getConfig().fragmentInURL ) {\n\t\t\tindex.f = undefined;\n\t\t}\n\n\t\t// If the current slide has an ID, use that as a named link,\n\t\t// but we don't support named links with a fragment index\n\t\tif( typeof id === 'string' && id.length ) {\n\t\t\turl = '/' + id;\n\n\t\t\t// If there is also a fragment, append that at the end\n\t\t\t// of the named link, like: #/named-link/2\n\t\t\tif( index.f >= 0 ) url += '/' + index.f;\n\t\t}\n\t\t// Otherwise use the /h/v index\n\t\telse {\n\t\t\tlet hashIndexBase = this.Reveal.getConfig().hashOneBasedIndex ? 1 : 0;\n\t\t\tif( index.h > 0 || index.v > 0 || index.f >= 0 ) url += index.h + hashIndexBase;\n\t\t\tif( index.v > 0 || index.f >= 0 ) url += '/' + (index.v + hashIndexBase );\n\t\t\tif( index.f >= 0 ) url += '/' + index.f;\n\t\t}\n\n\t\treturn url;\n\n\t}\n\n\t/**\n\t * Handler for the window level 'hashchange' event.\n\t *\n\t * @param {object} [event]\n\t */\n\tonWindowHashChange( event ) {\n\n\t\tthis.readURL();\n\n\t}\n\n}","import { queryAll } from '../utils/util.js'\nimport { isAndroid } from '../utils/device.js'\n\n/**\n * Manages our presentation controls. This includes both\n * the built-in control arrows as well as event monitoring\n * of any elements within the presentation with either of the\n * following helper classes:\n * - .navigate-up\n * - .navigate-right\n * - .navigate-down\n * - .navigate-left\n * - .navigate-next\n * - .navigate-prev\n */\nexport default class Controls {\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t\tthis.onNavigateLeftClicked = this.onNavigateLeftClicked.bind( this );\n\t\tthis.onNavigateRightClicked = this.onNavigateRightClicked.bind( this );\n\t\tthis.onNavigateUpClicked = this.onNavigateUpClicked.bind( this );\n\t\tthis.onNavigateDownClicked = this.onNavigateDownClicked.bind( this );\n\t\tthis.onNavigatePrevClicked = this.onNavigatePrevClicked.bind( this );\n\t\tthis.onNavigateNextClicked = this.onNavigateNextClicked.bind( this );\n\n\t}\n\n\trender() {\n\n\t\tconst rtl = this.Reveal.getConfig().rtl;\n\t\tconst revealElement = this.Reveal.getRevealElement();\n\n\t\tthis.element = document.createElement( 'aside' );\n\t\tthis.element.className = 'controls';\n\t\tthis.element.innerHTML =\n\t\t\t`\n\t\t\t\n\t\t\t\n\t\t\t`;\n\n\t\tthis.Reveal.getRevealElement().appendChild( this.element );\n\n\t\t// There can be multiple instances of controls throughout the page\n\t\tthis.controlsLeft = queryAll( revealElement, '.navigate-left' );\n\t\tthis.controlsRight = queryAll( revealElement, '.navigate-right' );\n\t\tthis.controlsUp = queryAll( revealElement, '.navigate-up' );\n\t\tthis.controlsDown = queryAll( revealElement, '.navigate-down' );\n\t\tthis.controlsPrev = queryAll( revealElement, '.navigate-prev' );\n\t\tthis.controlsNext = queryAll( revealElement, '.navigate-next' );\n\n\t\t// The left, right and down arrows in the standard reveal.js controls\n\t\tthis.controlsRightArrow = this.element.querySelector( '.navigate-right' );\n\t\tthis.controlsLeftArrow = this.element.querySelector( '.navigate-left' );\n\t\tthis.controlsDownArrow = this.element.querySelector( '.navigate-down' );\n\n\t}\n\n\t/**\n\t * Called when the reveal.js config is updated.\n\t */\n\tconfigure( config, oldConfig ) {\n\n\t\tthis.element.style.display = config.controls ? 'block' : 'none';\n\n\t\tthis.element.setAttribute( 'data-controls-layout', config.controlsLayout );\n\t\tthis.element.setAttribute( 'data-controls-back-arrows', config.controlsBackArrows );\n\n\t}\n\n\tbind() {\n\n\t\t// Listen to both touch and click events, in case the device\n\t\t// supports both\n\t\tlet pointerEvents = [ 'touchstart', 'click' ];\n\n\t\t// Only support touch for Android, fixes double navigations in\n\t\t// stock browser\n\t\tif( isAndroid ) {\n\t\t\tpointerEvents = [ 'touchstart' ];\n\t\t}\n\n\t\tpointerEvents.forEach( eventName => {\n\t\t\tthis.controlsLeft.forEach( el => el.addEventListener( eventName, this.onNavigateLeftClicked, false ) );\n\t\t\tthis.controlsRight.forEach( el => el.addEventListener( eventName, this.onNavigateRightClicked, false ) );\n\t\t\tthis.controlsUp.forEach( el => el.addEventListener( eventName, this.onNavigateUpClicked, false ) );\n\t\t\tthis.controlsDown.forEach( el => el.addEventListener( eventName, this.onNavigateDownClicked, false ) );\n\t\t\tthis.controlsPrev.forEach( el => el.addEventListener( eventName, this.onNavigatePrevClicked, false ) );\n\t\t\tthis.controlsNext.forEach( el => el.addEventListener( eventName, this.onNavigateNextClicked, false ) );\n\t\t} );\n\n\t}\n\n\tunbind() {\n\n\t\t[ 'touchstart', 'click' ].forEach( eventName => {\n\t\t\tthis.controlsLeft.forEach( el => el.removeEventListener( eventName, this.onNavigateLeftClicked, false ) );\n\t\t\tthis.controlsRight.forEach( el => el.removeEventListener( eventName, this.onNavigateRightClicked, false ) );\n\t\t\tthis.controlsUp.forEach( el => el.removeEventListener( eventName, this.onNavigateUpClicked, false ) );\n\t\t\tthis.controlsDown.forEach( el => el.removeEventListener( eventName, this.onNavigateDownClicked, false ) );\n\t\t\tthis.controlsPrev.forEach( el => el.removeEventListener( eventName, this.onNavigatePrevClicked, false ) );\n\t\t\tthis.controlsNext.forEach( el => el.removeEventListener( eventName, this.onNavigateNextClicked, false ) );\n\t\t} );\n\n\t}\n\n\t/**\n\t * Updates the state of all control/navigation arrows.\n\t */\n\tupdate() {\n\n\t\tlet routes = this.Reveal.availableRoutes();\n\n\t\t// Remove the 'enabled' class from all directions\n\t\t[...this.controlsLeft, ...this.controlsRight, ...this.controlsUp, ...this.controlsDown, ...this.controlsPrev, ...this.controlsNext].forEach( node => {\n\t\t\tnode.classList.remove( 'enabled', 'fragmented' );\n\n\t\t\t// Set 'disabled' attribute on all directions\n\t\t\tnode.setAttribute( 'disabled', 'disabled' );\n\t\t} );\n\n\t\t// Add the 'enabled' class to the available routes; remove 'disabled' attribute to enable buttons\n\t\tif( routes.left ) this.controlsLeft.forEach( el => { el.classList.add( 'enabled' ); el.removeAttribute( 'disabled' ); } );\n\t\tif( routes.right ) this.controlsRight.forEach( el => { el.classList.add( 'enabled' ); el.removeAttribute( 'disabled' ); } );\n\t\tif( routes.up ) this.controlsUp.forEach( el => { el.classList.add( 'enabled' ); el.removeAttribute( 'disabled' ); } );\n\t\tif( routes.down ) this.controlsDown.forEach( el => { el.classList.add( 'enabled' ); el.removeAttribute( 'disabled' ); } );\n\n\t\t// Prev/next buttons\n\t\tif( routes.left || routes.up ) this.controlsPrev.forEach( el => { el.classList.add( 'enabled' ); el.removeAttribute( 'disabled' ); } );\n\t\tif( routes.right || routes.down ) this.controlsNext.forEach( el => { el.classList.add( 'enabled' ); el.removeAttribute( 'disabled' ); } );\n\n\t\t// Highlight fragment directions\n\t\tlet currentSlide = this.Reveal.getCurrentSlide();\n\t\tif( currentSlide ) {\n\n\t\t\tlet fragmentsRoutes = this.Reveal.fragments.availableRoutes();\n\n\t\t\t// Always apply fragment decorator to prev/next buttons\n\t\t\tif( fragmentsRoutes.prev ) this.controlsPrev.forEach( el => { el.classList.add( 'fragmented', 'enabled' ); el.removeAttribute( 'disabled' ); } );\n\t\t\tif( fragmentsRoutes.next ) this.controlsNext.forEach( el => { el.classList.add( 'fragmented', 'enabled' ); el.removeAttribute( 'disabled' ); } );\n\n\t\t\t// Apply fragment decorators to directional buttons based on\n\t\t\t// what slide axis they are in\n\t\t\tif( this.Reveal.isVerticalSlide( currentSlide ) ) {\n\t\t\t\tif( fragmentsRoutes.prev ) this.controlsUp.forEach( el => { el.classList.add( 'fragmented', 'enabled' ); el.removeAttribute( 'disabled' ); } );\n\t\t\t\tif( fragmentsRoutes.next ) this.controlsDown.forEach( el => { el.classList.add( 'fragmented', 'enabled' ); el.removeAttribute( 'disabled' ); } );\n\t\t\t}\n\t\t\telse {\n\t\t\t\tif( fragmentsRoutes.prev ) this.controlsLeft.forEach( el => { el.classList.add( 'fragmented', 'enabled' ); el.removeAttribute( 'disabled' ); } );\n\t\t\t\tif( fragmentsRoutes.next ) this.controlsRight.forEach( el => { el.classList.add( 'fragmented', 'enabled' ); el.removeAttribute( 'disabled' ); } );\n\t\t\t}\n\n\t\t}\n\n\t\tif( this.Reveal.getConfig().controlsTutorial ) {\n\n\t\t\tlet indices = this.Reveal.getIndices();\n\n\t\t\t// Highlight control arrows with an animation to ensure\n\t\t\t// that the viewer knows how to navigate\n\t\t\tif( !this.Reveal.hasNavigatedVertically() && routes.down ) {\n\t\t\t\tthis.controlsDownArrow.classList.add( 'highlight' );\n\t\t\t}\n\t\t\telse {\n\t\t\t\tthis.controlsDownArrow.classList.remove( 'highlight' );\n\n\t\t\t\tif( this.Reveal.getConfig().rtl ) {\n\n\t\t\t\t\tif( !this.Reveal.hasNavigatedHorizontally() && routes.left && indices.v === 0 ) {\n\t\t\t\t\t\tthis.controlsLeftArrow.classList.add( 'highlight' );\n\t\t\t\t\t}\n\t\t\t\t\telse {\n\t\t\t\t\t\tthis.controlsLeftArrow.classList.remove( 'highlight' );\n\t\t\t\t\t}\n\n\t\t\t\t} else {\n\n\t\t\t\t\tif( !this.Reveal.hasNavigatedHorizontally() && routes.right && indices.v === 0 ) {\n\t\t\t\t\t\tthis.controlsRightArrow.classList.add( 'highlight' );\n\t\t\t\t\t}\n\t\t\t\t\telse {\n\t\t\t\t\t\tthis.controlsRightArrow.classList.remove( 'highlight' );\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\tdestroy() {\n\n\t\tthis.unbind();\n\t\tthis.element.remove();\n\n\t}\n\n\t/**\n\t * Event handlers for navigation control buttons.\n\t */\n\tonNavigateLeftClicked( event ) {\n\n\t\tevent.preventDefault();\n\t\tthis.Reveal.onUserInput();\n\n\t\tif( this.Reveal.getConfig().navigationMode === 'linear' ) {\n\t\t\tthis.Reveal.prev();\n\t\t}\n\t\telse {\n\t\t\tthis.Reveal.left();\n\t\t}\n\n\t}\n\n\tonNavigateRightClicked( event ) {\n\n\t\tevent.preventDefault();\n\t\tthis.Reveal.onUserInput();\n\n\t\tif( this.Reveal.getConfig().navigationMode === 'linear' ) {\n\t\t\tthis.Reveal.next();\n\t\t}\n\t\telse {\n\t\t\tthis.Reveal.right();\n\t\t}\n\n\t}\n\n\tonNavigateUpClicked( event ) {\n\n\t\tevent.preventDefault();\n\t\tthis.Reveal.onUserInput();\n\n\t\tthis.Reveal.up();\n\n\t}\n\n\tonNavigateDownClicked( event ) {\n\n\t\tevent.preventDefault();\n\t\tthis.Reveal.onUserInput();\n\n\t\tthis.Reveal.down();\n\n\t}\n\n\tonNavigatePrevClicked( event ) {\n\n\t\tevent.preventDefault();\n\t\tthis.Reveal.onUserInput();\n\n\t\tthis.Reveal.prev();\n\n\t}\n\n\tonNavigateNextClicked( event ) {\n\n\t\tevent.preventDefault();\n\t\tthis.Reveal.onUserInput();\n\n\t\tthis.Reveal.next();\n\n\t}\n\n\n}","/**\n * Creates a visual progress bar for the presentation.\n */\nexport default class Progress {\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t\tthis.onProgressClicked = this.onProgressClicked.bind( this );\n\n\t}\n\n\trender() {\n\n\t\tthis.element = document.createElement( 'div' );\n\t\tthis.element.className = 'progress';\n\t\tthis.Reveal.getRevealElement().appendChild( this.element );\n\n\t\tthis.bar = document.createElement( 'span' );\n\t\tthis.element.appendChild( this.bar );\n\n\t}\n\n\t/**\n\t * Called when the reveal.js config is updated.\n\t */\n\tconfigure( config, oldConfig ) {\n\n\t\tthis.element.style.display = config.progress ? 'block' : 'none';\n\n\t}\n\n\tbind() {\n\n\t\tif( this.Reveal.getConfig().progress && this.element ) {\n\t\t\tthis.element.addEventListener( 'click', this.onProgressClicked, false );\n\t\t}\n\n\t}\n\n\tunbind() {\n\n\t\tif ( this.Reveal.getConfig().progress && this.element ) {\n\t\t\tthis.element.removeEventListener( 'click', this.onProgressClicked, false );\n\t\t}\n\n\t}\n\n\t/**\n\t * Updates the progress bar to reflect the current slide.\n\t */\n\tupdate() {\n\n\t\t// Update progress if enabled\n\t\tif( this.Reveal.getConfig().progress && this.bar ) {\n\n\t\t\tlet scale = this.Reveal.getProgress();\n\n\t\t\t// Don't fill the progress bar if there's only one slide\n\t\t\tif( this.Reveal.getTotalSlides() < 2 ) {\n\t\t\t\tscale = 0;\n\t\t\t}\n\n\t\t\tthis.bar.style.transform = 'scaleX('+ scale +')';\n\n\t\t}\n\n\t}\n\n\tgetMaxWidth() {\n\n\t\treturn this.Reveal.getRevealElement().offsetWidth;\n\n\t}\n\n\t/**\n\t * Clicking on the progress bar results in a navigation to the\n\t * closest approximate horizontal slide using this equation:\n\t *\n\t * ( clickX / presentationWidth ) * numberOfSlides\n\t *\n\t * @param {object} event\n\t */\n\tonProgressClicked( event ) {\n\n\t\tthis.Reveal.onUserInput( event );\n\n\t\tevent.preventDefault();\n\n\t\tlet slides = this.Reveal.getSlides();\n\t\tlet slidesTotal = slides.length;\n\t\tlet slideIndex = Math.floor( ( event.clientX / this.getMaxWidth() ) * slidesTotal );\n\n\t\tif( this.Reveal.getConfig().rtl ) {\n\t\t\tslideIndex = slidesTotal - slideIndex;\n\t\t}\n\n\t\tlet targetIndices = this.Reveal.getIndices(slides[slideIndex]);\n\t\tthis.Reveal.slide( targetIndices.h, targetIndices.v );\n\n\t}\n\n\tdestroy() {\n\n\t\tthis.element.remove();\n\n\t}\n\n}","/**\n * Handles hiding of the pointer/cursor when inactive.\n */\nexport default class Pointer {\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t\t// Throttles mouse wheel navigation\n\t\tthis.lastMouseWheelStep = 0;\n\n\t\t// Is the mouse pointer currently hidden from view\n\t\tthis.cursorHidden = false;\n\n\t\t// Timeout used to determine when the cursor is inactive\n\t\tthis.cursorInactiveTimeout = 0;\n\n\t\tthis.onDocumentCursorActive = this.onDocumentCursorActive.bind( this );\n\t\tthis.onDocumentMouseScroll = this.onDocumentMouseScroll.bind( this );\n\n\t}\n\n\t/**\n\t * Called when the reveal.js config is updated.\n\t */\n\tconfigure( config, oldConfig ) {\n\n\t\tif( config.mouseWheel ) {\n\t\t\tdocument.addEventListener( 'DOMMouseScroll', this.onDocumentMouseScroll, false ); // FF\n\t\t\tdocument.addEventListener( 'mousewheel', this.onDocumentMouseScroll, false );\n\t\t}\n\t\telse {\n\t\t\tdocument.removeEventListener( 'DOMMouseScroll', this.onDocumentMouseScroll, false ); // FF\n\t\t\tdocument.removeEventListener( 'mousewheel', this.onDocumentMouseScroll, false );\n\t\t}\n\n\t\t// Auto-hide the mouse pointer when its inactive\n\t\tif( config.hideInactiveCursor ) {\n\t\t\tdocument.addEventListener( 'mousemove', this.onDocumentCursorActive, false );\n\t\t\tdocument.addEventListener( 'mousedown', this.onDocumentCursorActive, false );\n\t\t}\n\t\telse {\n\t\t\tthis.showCursor();\n\n\t\t\tdocument.removeEventListener( 'mousemove', this.onDocumentCursorActive, false );\n\t\t\tdocument.removeEventListener( 'mousedown', this.onDocumentCursorActive, false );\n\t\t}\n\n\t}\n\n\t/**\n\t * Shows the mouse pointer after it has been hidden with\n\t * #hideCursor.\n\t */\n\tshowCursor() {\n\n\t\tif( this.cursorHidden ) {\n\t\t\tthis.cursorHidden = false;\n\t\t\tthis.Reveal.getRevealElement().style.cursor = '';\n\t\t}\n\n\t}\n\n\t/**\n\t * Hides the mouse pointer when it's on top of the .reveal\n\t * container.\n\t */\n\thideCursor() {\n\n\t\tif( this.cursorHidden === false ) {\n\t\t\tthis.cursorHidden = true;\n\t\t\tthis.Reveal.getRevealElement().style.cursor = 'none';\n\t\t}\n\n\t}\n\n\tdestroy() {\n\n\t\tthis.showCursor();\n\n\t\tdocument.removeEventListener( 'DOMMouseScroll', this.onDocumentMouseScroll, false );\n\t\tdocument.removeEventListener( 'mousewheel', this.onDocumentMouseScroll, false );\n\t\tdocument.removeEventListener( 'mousemove', this.onDocumentCursorActive, false );\n\t\tdocument.removeEventListener( 'mousedown', this.onDocumentCursorActive, false );\n\n\t}\n\n\t/**\n\t * Called whenever there is mouse input at the document level\n\t * to determine if the cursor is active or not.\n\t *\n\t * @param {object} event\n\t */\n\tonDocumentCursorActive( event ) {\n\n\t\tthis.showCursor();\n\n\t\tclearTimeout( this.cursorInactiveTimeout );\n\n\t\tthis.cursorInactiveTimeout = setTimeout( this.hideCursor.bind( this ), this.Reveal.getConfig().hideCursorTime );\n\n\t}\n\n\t/**\n\t * Handles mouse wheel scrolling, throttled to avoid skipping\n\t * multiple slides.\n\t *\n\t * @param {object} event\n\t */\n\tonDocumentMouseScroll( event ) {\n\n\t\tif( Date.now() - this.lastMouseWheelStep > 1000 ) {\n\n\t\t\tthis.lastMouseWheelStep = Date.now();\n\n\t\t\tlet delta = event.detail || -event.wheelDelta;\n\t\t\tif( delta > 0 ) {\n\t\t\t\tthis.Reveal.next();\n\t\t\t}\n\t\t\telse if( delta < 0 ) {\n\t\t\t\tthis.Reveal.prev();\n\t\t\t}\n\n\t\t}\n\n\t}\n\n}","/**\n * Loads a JavaScript file from the given URL and executes it.\n *\n * @param {string} url Address of the .js file to load\n * @param {function} callback Method to invoke when the script\n * has loaded and executed\n */\nexport const loadScript = ( url, callback ) => {\n\n\tconst script = document.createElement( 'script' );\n\tscript.type = 'text/javascript';\n\tscript.async = false;\n\tscript.defer = false;\n\tscript.src = url;\n\n\tif( typeof callback === 'function' ) {\n\n\t\t// Success callback\n\t\tscript.onload = script.onreadystatechange = event => {\n\t\t\tif( event.type === 'load' || /loaded|complete/.test( script.readyState ) ) {\n\n\t\t\t\t// Kill event listeners\n\t\t\t\tscript.onload = script.onreadystatechange = script.onerror = null;\n\n\t\t\t\tcallback();\n\n\t\t\t}\n\t\t};\n\n\t\t// Error callback\n\t\tscript.onerror = err => {\n\n\t\t\t// Kill event listeners\n\t\t\tscript.onload = script.onreadystatechange = script.onerror = null;\n\n\t\t\tcallback( new Error( 'Failed loading script: ' + script.src + '\\n' + err ) );\n\n\t\t};\n\n\t}\n\n\t// Append the script at the end of \n\tconst head = document.querySelector( 'head' );\n\thead.insertBefore( script, head.lastChild );\n\n}","import { loadScript } from '../utils/loader.js'\n\n/**\n * Manages loading and registering of reveal.js plugins.\n */\nexport default class Plugins {\n\n\tconstructor( reveal ) {\n\n\t\tthis.Reveal = reveal;\n\n\t\t// Flags our current state (idle -> loading -> loaded)\n\t\tthis.state = 'idle';\n\n\t\t// An id:instance map of currently registed plugins\n\t\tthis.registeredPlugins = {};\n\n\t\tthis.asyncDependencies = [];\n\n\t}\n\n\t/**\n\t * Loads reveal.js dependencies, registers and\n\t * initializes plugins.\n\t *\n\t * Plugins are direct references to a reveal.js plugin\n\t * object that we register and initialize after any\n\t * synchronous dependencies have loaded.\n\t *\n\t * Dependencies are defined via the 'dependencies' config\n\t * option and will be loaded prior to starting reveal.js.\n\t * Some dependencies may have an 'async' flag, if so they\n\t * will load after reveal.js has been started up.\n\t */\n\tload( plugins, dependencies ) {\n\n\t\tthis.state = 'loading';\n\n\t\tplugins.forEach( this.registerPlugin.bind( this ) );\n\n\t\treturn new Promise( resolve => {\n\n\t\t\tlet scripts = [],\n\t\t\t\tscriptsToLoad = 0;\n\n\t\t\tdependencies.forEach( s => {\n\t\t\t\t// Load if there's no condition or the condition is truthy\n\t\t\t\tif( !s.condition || s.condition() ) {\n\t\t\t\t\tif( s.async ) {\n\t\t\t\t\t\tthis.asyncDependencies.push( s );\n\t\t\t\t\t}\n\t\t\t\t\telse {\n\t\t\t\t\t\tscripts.push( s );\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t} );\n\n\t\t\tif( scripts.length ) {\n\t\t\t\tscriptsToLoad = scripts.length;\n\n\t\t\t\tconst scriptLoadedCallback = (s) => {\n\t\t\t\t\tif( s && typeof s.callback === 'function' ) s.callback();\n\n\t\t\t\t\tif( --scriptsToLoad === 0 ) {\n\t\t\t\t\t\tthis.initPlugins().then( resolve );\n\t\t\t\t\t}\n\t\t\t\t};\n\n\t\t\t\t// Load synchronous scripts\n\t\t\t\tscripts.forEach( s => {\n\t\t\t\t\tif( typeof s.id === 'string' ) {\n\t\t\t\t\t\tthis.registerPlugin( s );\n\t\t\t\t\t\tscriptLoadedCallback( s );\n\t\t\t\t\t}\n\t\t\t\t\telse if( typeof s.src === 'string' ) {\n\t\t\t\t\t\tloadScript( s.src, () => scriptLoadedCallback(s) );\n\t\t\t\t\t}\n\t\t\t\t\telse {\n\t\t\t\t\t\tconsole.warn( 'Unrecognized plugin format', s );\n\t\t\t\t\t\tscriptLoadedCallback();\n\t\t\t\t\t}\n\t\t\t\t} );\n\t\t\t}\n\t\t\telse {\n\t\t\t\tthis.initPlugins().then( resolve );\n\t\t\t}\n\n\t\t} );\n\n\t}\n\n\t/**\n\t * Initializes our plugins and waits for them to be ready\n\t * before proceeding.\n\t */\n\tinitPlugins() {\n\n\t\treturn new Promise( resolve => {\n\n\t\t\tlet pluginValues = Object.values( this.registeredPlugins );\n\t\t\tlet pluginsToInitialize = pluginValues.length;\n\n\t\t\t// If there are no plugins, skip this step\n\t\t\tif( pluginsToInitialize === 0 ) {\n\t\t\t\tthis.loadAsync().then( resolve );\n\t\t\t}\n\t\t\t// ... otherwise initialize plugins\n\t\t\telse {\n\n\t\t\t\tlet initNextPlugin;\n\n\t\t\t\tlet afterPlugInitialized = () => {\n\t\t\t\t\tif( --pluginsToInitialize === 0 ) {\n\t\t\t\t\t\tthis.loadAsync().then( resolve );\n\t\t\t\t\t}\n\t\t\t\t\telse {\n\t\t\t\t\t\tinitNextPlugin();\n\t\t\t\t\t}\n\t\t\t\t};\n\n\t\t\t\tlet i = 0;\n\n\t\t\t\t// Initialize plugins serially\n\t\t\t\tinitNextPlugin = () => {\n\n\t\t\t\t\tlet plugin = pluginValues[i++];\n\n\t\t\t\t\t// If the plugin has an 'init' method, invoke it\n\t\t\t\t\tif( typeof plugin.init === 'function' ) {\n\t\t\t\t\t\tlet promise = plugin.init( this.Reveal );\n\n\t\t\t\t\t\t// If the plugin returned a Promise, wait for it\n\t\t\t\t\t\tif( promise && typeof promise.then === 'function' ) {\n\t\t\t\t\t\t\tpromise.then( afterPlugInitialized );\n\t\t\t\t\t\t}\n\t\t\t\t\t\telse {\n\t\t\t\t\t\t\tafterPlugInitialized();\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\telse {\n\t\t\t\t\t\tafterPlugInitialized();\n\t\t\t\t\t}\n\n\t\t\t\t}\n\n\t\t\t\tinitNextPlugin();\n\n\t\t\t}\n\n\t\t} )\n\n\t}\n\n\t/**\n\t * Loads all async reveal.js dependencies.\n\t */\n\tloadAsync() {\n\n\t\tthis.state = 'loaded';\n\n\t\tif( this.asyncDependencies.length ) {\n\t\t\tthis.asyncDependencies.forEach( s => {\n\t\t\t\tloadScript( s.src, s.callback );\n\t\t\t} );\n\t\t}\n\n\t\treturn Promise.resolve();\n\n\t}\n\n\t/**\n\t * Registers a new plugin with this reveal.js instance.\n\t *\n\t * reveal.js waits for all regisered plugins to initialize\n\t * before considering itself ready, as long as the plugin\n\t * is registered before calling `Reveal.initialize()`.\n\t */\n\tregisterPlugin( plugin ) {\n\n\t\t// Backwards compatibility to make reveal.js ~3.9.0\n\t\t// plugins work with reveal.js 4.0.0\n\t\tif( arguments.length === 2 && typeof arguments[0] === 'string' ) {\n\t\t\tplugin = arguments[1];\n\t\t\tplugin.id = arguments[0];\n\t\t}\n\t\t// Plugin can optionally be a function which we call\n\t\t// to create an instance of the plugin\n\t\telse if( typeof plugin === 'function' ) {\n\t\t\tplugin = plugin();\n\t\t}\n\n\t\tlet id = plugin.id;\n\n\t\tif( typeof id !== 'string' ) {\n\t\t\tconsole.warn( 'Unrecognized plugin format; can\\'t find plugin.id', plugin );\n\t\t}\n\t\telse if( this.registeredPlugins[id] === undefined ) {\n\t\t\tthis.registeredPlugins[id] = plugin;\n\n\t\t\t// If a plugin is registered after reveal.js is loaded,\n\t\t\t// initialize it right away\n\t\t\tif( this.state === 'loaded' && typeof plugin.init === 'function' ) {\n\t\t\t\tplugin.init( this.Reveal );\n\t\t\t}\n\t\t}\n\t\telse {\n\t\t\tconsole.warn( 'reveal.js: \"'+ id +'\" plugin has already been registered' );\n\t\t}\n\n\t}\n\n\t/**\n\t * Checks if a specific plugin has been registered.\n\t *\n\t * @param {String} id Unique plugin identifier\n\t */\n\thasPlugin( id ) {\n\n\t\treturn !!this.registeredPlugins[id];\n\n\t}\n\n\t/**\n\t * Returns the specific plugin instance, if a plugin\n\t * with the given ID has been registered.\n\t *\n\t * @param {String} id Unique plugin identifier\n\t */\n\tgetPlugin( id ) {\n\n\t\treturn this.registeredPlugins[id];\n\n\t}\n\n\tgetRegisteredPlugins() {\n\n\t\treturn this.registeredPlugins;\n\n\t}\n\n\tdestroy() {\n\n\t\tObject.values( this.registeredPlugins ).forEach( plugin => {\n\t\t\tif( typeof plugin.destroy === 'function' ) {\n\t\t\t\tplugin.destroy();\n\t\t\t}\n\t\t} );\n\n\t\tthis.registeredPlugins = {};\n\t\tthis.asyncDependencies = [];\n\n\t}\n\n}\n","import { SLIDES_SELECTOR } from '../utils/constants.js'\nimport { queryAll, createStyleSheet } from '../utils/util.js'\n\n/**\n * Setups up our presentation for printing/exporting to PDF.\n */\nexport default class Print {\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t}\n\n\t/**\n\t * Configures the presentation for printing to a static\n\t * PDF.\n\t */\n\tasync setupPDF() {\n\n\t\tconst config = this.Reveal.getConfig();\n\t\tconst slides = queryAll( this.Reveal.getRevealElement(), SLIDES_SELECTOR )\n\n\t\t// Compute slide numbers now, before we start duplicating slides\n\t\tconst doingSlideNumbers = config.slideNumber && /all|print/i.test( config.showSlideNumber );\n\n\t\tconst slideSize = this.Reveal.getComputedSlideSize( window.innerWidth, window.innerHeight );\n\n\t\t// Dimensions of the PDF pages\n\t\tconst pageWidth = Math.floor( slideSize.width * ( 1 + config.margin ) ),\n\t\t\tpageHeight = Math.floor( slideSize.height * ( 1 + config.margin ) );\n\n\t\t// Dimensions of slides within the pages\n\t\tconst slideWidth = slideSize.width,\n\t\t\tslideHeight = slideSize.height;\n\n\t\tawait new Promise( requestAnimationFrame );\n\n\t\t// Let the browser know what page size we want to print\n\t\tcreateStyleSheet( '@page{size:'+ pageWidth +'px '+ pageHeight +'px; margin: 0px;}' );\n\n\t\t// Limit the size of certain elements to the dimensions of the slide\n\t\tcreateStyleSheet( '.reveal section>img, .reveal section>video, .reveal section>iframe{max-width: '+ slideWidth +'px; max-height:'+ slideHeight +'px}' );\n\n\t\tdocument.documentElement.classList.add( 'print-pdf' );\n\t\tdocument.body.style.width = pageWidth + 'px';\n\t\tdocument.body.style.height = pageHeight + 'px';\n\n\t\tconst viewportElement = document.querySelector( '.reveal-viewport' );\n\t\tlet presentationBackground;\n\t\tif( viewportElement ) {\n\t\t\tconst viewportStyles = window.getComputedStyle( viewportElement );\n\t\t\tif( viewportStyles && viewportStyles.background ) {\n\t\t\t\tpresentationBackground = viewportStyles.background;\n\t\t\t}\n\t\t}\n\n\t\t// Make sure stretch elements fit on slide\n\t\tawait new Promise( requestAnimationFrame );\n\t\tthis.Reveal.layoutSlideContents( slideWidth, slideHeight );\n\n\t\t// Batch scrollHeight access to prevent layout thrashing\n\t\tawait new Promise( requestAnimationFrame );\n\n\t\tconst slideScrollHeights = slides.map( slide => slide.scrollHeight );\n\n\t\tconst pages = [];\n\t\tconst pageContainer = slides[0].parentNode;\n\n\t\t// Slide and slide background layout\n\t\tslides.forEach( function( slide, index ) {\n\n\t\t\t// Vertical stacks are not centred since their section\n\t\t\t// children will be\n\t\t\tif( slide.classList.contains( 'stack' ) === false ) {\n\t\t\t\t// Center the slide inside of the page, giving the slide some margin\n\t\t\t\tlet left = ( pageWidth - slideWidth ) / 2;\n\t\t\t\tlet top = ( pageHeight - slideHeight ) / 2;\n\n\t\t\t\tconst contentHeight = slideScrollHeights[ index ];\n\t\t\t\tlet numberOfPages = Math.max( Math.ceil( contentHeight / pageHeight ), 1 );\n\n\t\t\t\t// Adhere to configured pages per slide limit\n\t\t\t\tnumberOfPages = Math.min( numberOfPages, config.pdfMaxPagesPerSlide );\n\n\t\t\t\t// Center slides vertically\n\t\t\t\tif( numberOfPages === 1 && config.center || slide.classList.contains( 'center' ) ) {\n\t\t\t\t\ttop = Math.max( ( pageHeight - contentHeight ) / 2, 0 );\n\t\t\t\t}\n\n\t\t\t\t// Wrap the slide in a page element and hide its overflow\n\t\t\t\t// so that no page ever flows onto another\n\t\t\t\tconst page = document.createElement( 'div' );\n\t\t\t\tpages.push( page );\n\n\t\t\t\tpage.className = 'pdf-page';\n\t\t\t\tpage.style.height = ( ( pageHeight + config.pdfPageHeightOffset ) * numberOfPages ) + 'px';\n\n\t\t\t\t// Copy the presentation-wide background to each individual\n\t\t\t\t// page when printing\n\t\t\t\tif( presentationBackground ) {\n\t\t\t\t\tpage.style.background = presentationBackground;\n\t\t\t\t}\n\n\t\t\t\tpage.appendChild( slide );\n\n\t\t\t\t// Position the slide inside of the page\n\t\t\t\tslide.style.left = left + 'px';\n\t\t\t\tslide.style.top = top + 'px';\n\t\t\t\tslide.style.width = slideWidth + 'px';\n\n\t\t\t\t// Re-run the slide layout so that r-fit-text is applied based on\n\t\t\t\t// the printed slide size\n\t\t\t\tthis.Reveal.slideContent.layout( slide )\n\n\t\t\t\tif( slide.slideBackgroundElement ) {\n\t\t\t\t\tpage.insertBefore( slide.slideBackgroundElement, slide );\n\t\t\t\t}\n\n\t\t\t\t// Inject notes if `showNotes` is enabled\n\t\t\t\tif( config.showNotes ) {\n\n\t\t\t\t\t// Are there notes for this slide?\n\t\t\t\t\tconst notes = this.Reveal.getSlideNotes( slide );\n\t\t\t\t\tif( notes ) {\n\n\t\t\t\t\t\tconst notesSpacing = 8;\n\t\t\t\t\t\tconst notesLayout = typeof config.showNotes === 'string' ? config.showNotes : 'inline';\n\t\t\t\t\t\tconst notesElement = document.createElement( 'div' );\n\t\t\t\t\t\tnotesElement.classList.add( 'speaker-notes' );\n\t\t\t\t\t\tnotesElement.classList.add( 'speaker-notes-pdf' );\n\t\t\t\t\t\tnotesElement.setAttribute( 'data-layout', notesLayout );\n\t\t\t\t\t\tnotesElement.innerHTML = notes;\n\n\t\t\t\t\t\tif( notesLayout === 'separate-page' ) {\n\t\t\t\t\t\t\tpages.push( notesElement );\n\t\t\t\t\t\t}\n\t\t\t\t\t\telse {\n\t\t\t\t\t\t\tnotesElement.style.left = notesSpacing + 'px';\n\t\t\t\t\t\t\tnotesElement.style.bottom = notesSpacing + 'px';\n\t\t\t\t\t\t\tnotesElement.style.width = ( pageWidth - notesSpacing*2 ) + 'px';\n\t\t\t\t\t\t\tpage.appendChild( notesElement );\n\t\t\t\t\t\t}\n\n\t\t\t\t\t}\n\n\t\t\t\t}\n\n\t\t\t\t// Inject slide numbers if `slideNumbers` are enabled\n\t\t\t\tif( doingSlideNumbers ) {\n\t\t\t\t\tconst slideNumber = index + 1;\n\t\t\t\t\tconst numberElement = document.createElement( 'div' );\n\t\t\t\t\tnumberElement.classList.add( 'slide-number' );\n\t\t\t\t\tnumberElement.classList.add( 'slide-number-pdf' );\n\t\t\t\t\tnumberElement.innerHTML = slideNumber;\n\t\t\t\t\tpage.appendChild( numberElement );\n\t\t\t\t}\n\n\t\t\t\t// Copy page and show fragments one after another\n\t\t\t\tif( config.pdfSeparateFragments ) {\n\n\t\t\t\t\t// Each fragment 'group' is an array containing one or more\n\t\t\t\t\t// fragments. Multiple fragments that appear at the same time\n\t\t\t\t\t// are part of the same group.\n\t\t\t\t\tconst fragmentGroups = this.Reveal.fragments.sort( page.querySelectorAll( '.fragment' ), true );\n\n\t\t\t\t\tlet previousFragmentStep;\n\n\t\t\t\t\tfragmentGroups.forEach( function( fragments ) {\n\n\t\t\t\t\t\t// Remove 'current-fragment' from the previous group\n\t\t\t\t\t\tif( previousFragmentStep ) {\n\t\t\t\t\t\t\tpreviousFragmentStep.forEach( function( fragment ) {\n\t\t\t\t\t\t\t\tfragment.classList.remove( 'current-fragment' );\n\t\t\t\t\t\t\t} );\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// Show the fragments for the current index\n\t\t\t\t\t\tfragments.forEach( function( fragment ) {\n\t\t\t\t\t\t\tfragment.classList.add( 'visible', 'current-fragment' );\n\t\t\t\t\t\t}, this );\n\n\t\t\t\t\t\t// Create a separate page for the current fragment state\n\t\t\t\t\t\tconst clonedPage = page.cloneNode( true );\n\t\t\t\t\t\tpages.push( clonedPage );\n\n\t\t\t\t\t\tpreviousFragmentStep = fragments;\n\n\t\t\t\t\t}, this );\n\n\t\t\t\t\t// Reset the first/original page so that all fragments are hidden\n\t\t\t\t\tfragmentGroups.forEach( function( fragments ) {\n\t\t\t\t\t\tfragments.forEach( function( fragment ) {\n\t\t\t\t\t\t\tfragment.classList.remove( 'visible', 'current-fragment' );\n\t\t\t\t\t\t} );\n\t\t\t\t\t} );\n\n\t\t\t\t}\n\t\t\t\t// Show all fragments\n\t\t\t\telse {\n\t\t\t\t\tqueryAll( page, '.fragment:not(.fade-out)' ).forEach( function( fragment ) {\n\t\t\t\t\t\tfragment.classList.add( 'visible' );\n\t\t\t\t\t} );\n\t\t\t\t}\n\n\t\t\t}\n\n\t\t}, this );\n\n\t\tawait new Promise( requestAnimationFrame );\n\n\t\tpages.forEach( page => pageContainer.appendChild( page ) );\n\n\t\t// Notify subscribers that the PDF layout is good to go\n\t\tthis.Reveal.dispatchEvent({ type: 'pdf-ready' });\n\n\t}\n\n\t/**\n\t * Checks if this instance is being used to print a PDF.\n\t */\n\tisPrintingPDF() {\n\n\t\treturn ( /print-pdf/gi ).test( window.location.search );\n\n\t}\n\n}\n","import { isAndroid } from '../utils/device.js'\nimport { matches } from '../utils/util.js'\n\nconst SWIPE_THRESHOLD = 40;\n\n/**\n * Controls all touch interactions and navigations for\n * a presentation.\n */\nexport default class Touch {\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t\t// Holds information about the currently ongoing touch interaction\n\t\tthis.touchStartX = 0;\n\t\tthis.touchStartY = 0;\n\t\tthis.touchStartCount = 0;\n\t\tthis.touchCaptured = false;\n\n\t\tthis.onPointerDown = this.onPointerDown.bind( this );\n\t\tthis.onPointerMove = this.onPointerMove.bind( this );\n\t\tthis.onPointerUp = this.onPointerUp.bind( this );\n\t\tthis.onTouchStart = this.onTouchStart.bind( this );\n\t\tthis.onTouchMove = this.onTouchMove.bind( this );\n\t\tthis.onTouchEnd = this.onTouchEnd.bind( this );\n\n\t}\n\n\t/**\n\t *\n\t */\n\tbind() {\n\n\t\tlet revealElement = this.Reveal.getRevealElement();\n\n\t\tif( 'onpointerdown' in window ) {\n\t\t\t// Use W3C pointer events\n\t\t\trevealElement.addEventListener( 'pointerdown', this.onPointerDown, false );\n\t\t\trevealElement.addEventListener( 'pointermove', this.onPointerMove, false );\n\t\t\trevealElement.addEventListener( 'pointerup', this.onPointerUp, false );\n\t\t}\n\t\telse if( window.navigator.msPointerEnabled ) {\n\t\t\t// IE 10 uses prefixed version of pointer events\n\t\t\trevealElement.addEventListener( 'MSPointerDown', this.onPointerDown, false );\n\t\t\trevealElement.addEventListener( 'MSPointerMove', this.onPointerMove, false );\n\t\t\trevealElement.addEventListener( 'MSPointerUp', this.onPointerUp, false );\n\t\t}\n\t\telse {\n\t\t\t// Fall back to touch events\n\t\t\trevealElement.addEventListener( 'touchstart', this.onTouchStart, false );\n\t\t\trevealElement.addEventListener( 'touchmove', this.onTouchMove, false );\n\t\t\trevealElement.addEventListener( 'touchend', this.onTouchEnd, false );\n\t\t}\n\n\t}\n\n\t/**\n\t *\n\t */\n\tunbind() {\n\n\t\tlet revealElement = this.Reveal.getRevealElement();\n\n\t\trevealElement.removeEventListener( 'pointerdown', this.onPointerDown, false );\n\t\trevealElement.removeEventListener( 'pointermove', this.onPointerMove, false );\n\t\trevealElement.removeEventListener( 'pointerup', this.onPointerUp, false );\n\n\t\trevealElement.removeEventListener( 'MSPointerDown', this.onPointerDown, false );\n\t\trevealElement.removeEventListener( 'MSPointerMove', this.onPointerMove, false );\n\t\trevealElement.removeEventListener( 'MSPointerUp', this.onPointerUp, false );\n\n\t\trevealElement.removeEventListener( 'touchstart', this.onTouchStart, false );\n\t\trevealElement.removeEventListener( 'touchmove', this.onTouchMove, false );\n\t\trevealElement.removeEventListener( 'touchend', this.onTouchEnd, false );\n\n\t}\n\n\t/**\n\t * Checks if the target element prevents the triggering of\n\t * swipe navigation.\n\t */\n\tisSwipePrevented( target ) {\n\n\t\t// Prevent accidental swipes when scrubbing timelines\n\t\tif( matches( target, 'video, audio' ) ) return true;\n\n\t\twhile( target && typeof target.hasAttribute === 'function' ) {\n\t\t\tif( target.hasAttribute( 'data-prevent-swipe' ) ) return true;\n\t\t\ttarget = target.parentNode;\n\t\t}\n\n\t\treturn false;\n\n\t}\n\n\t/**\n\t * Handler for the 'touchstart' event, enables support for\n\t * swipe and pinch gestures.\n\t *\n\t * @param {object} event\n\t */\n\tonTouchStart( event ) {\n\n\t\tif( this.isSwipePrevented( event.target ) ) return true;\n\n\t\tthis.touchStartX = event.touches[0].clientX;\n\t\tthis.touchStartY = event.touches[0].clientY;\n\t\tthis.touchStartCount = event.touches.length;\n\n\t}\n\n\t/**\n\t * Handler for the 'touchmove' event.\n\t *\n\t * @param {object} event\n\t */\n\tonTouchMove( event ) {\n\n\t\tif( this.isSwipePrevented( event.target ) ) return true;\n\n\t\tlet config = this.Reveal.getConfig();\n\n\t\t// Each touch should only trigger one action\n\t\tif( !this.touchCaptured ) {\n\t\t\tthis.Reveal.onUserInput( event );\n\n\t\t\tlet currentX = event.touches[0].clientX;\n\t\t\tlet currentY = event.touches[0].clientY;\n\n\t\t\t// There was only one touch point, look for a swipe\n\t\t\tif( event.touches.length === 1 && this.touchStartCount !== 2 ) {\n\n\t\t\t\tlet availableRoutes = this.Reveal.availableRoutes({ includeFragments: true });\n\n\t\t\t\tlet deltaX = currentX - this.touchStartX,\n\t\t\t\t\tdeltaY = currentY - this.touchStartY;\n\n\t\t\t\tif( deltaX > SWIPE_THRESHOLD && Math.abs( deltaX ) > Math.abs( deltaY ) ) {\n\t\t\t\t\tthis.touchCaptured = true;\n\t\t\t\t\tif( config.navigationMode === 'linear' ) {\n\t\t\t\t\t\tif( config.rtl ) {\n\t\t\t\t\t\t\tthis.Reveal.next();\n\t\t\t\t\t\t}\n\t\t\t\t\t\telse {\n\t\t\t\t\t\t\tthis.Reveal.prev();\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\telse {\n\t\t\t\t\t\tthis.Reveal.left();\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\telse if( deltaX < -SWIPE_THRESHOLD && Math.abs( deltaX ) > Math.abs( deltaY ) ) {\n\t\t\t\t\tthis.touchCaptured = true;\n\t\t\t\t\tif( config.navigationMode === 'linear' ) {\n\t\t\t\t\t\tif( config.rtl ) {\n\t\t\t\t\t\t\tthis.Reveal.prev();\n\t\t\t\t\t\t}\n\t\t\t\t\t\telse {\n\t\t\t\t\t\t\tthis.Reveal.next();\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\telse {\n\t\t\t\t\t\tthis.Reveal.right();\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\telse if( deltaY > SWIPE_THRESHOLD && availableRoutes.up ) {\n\t\t\t\t\tthis.touchCaptured = true;\n\t\t\t\t\tif( config.navigationMode === 'linear' ) {\n\t\t\t\t\t\tthis.Reveal.prev();\n\t\t\t\t\t}\n\t\t\t\t\telse {\n\t\t\t\t\t\tthis.Reveal.up();\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\telse if( deltaY < -SWIPE_THRESHOLD && availableRoutes.down ) {\n\t\t\t\t\tthis.touchCaptured = true;\n\t\t\t\t\tif( config.navigationMode === 'linear' ) {\n\t\t\t\t\t\tthis.Reveal.next();\n\t\t\t\t\t}\n\t\t\t\t\telse {\n\t\t\t\t\t\tthis.Reveal.down();\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\t// If we're embedded, only block touch events if they have\n\t\t\t\t// triggered an action\n\t\t\t\tif( config.embedded ) {\n\t\t\t\t\tif( this.touchCaptured || this.Reveal.isVerticalSlide() ) {\n\t\t\t\t\t\tevent.preventDefault();\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\t// Not embedded? Block them all to avoid needless tossing\n\t\t\t\t// around of the viewport in iOS\n\t\t\t\telse {\n\t\t\t\t\tevent.preventDefault();\n\t\t\t\t}\n\n\t\t\t}\n\t\t}\n\t\t// There's a bug with swiping on some Android devices unless\n\t\t// the default action is always prevented\n\t\telse if( isAndroid ) {\n\t\t\tevent.preventDefault();\n\t\t}\n\n\t}\n\n\t/**\n\t * Handler for the 'touchend' event.\n\t *\n\t * @param {object} event\n\t */\n\tonTouchEnd( event ) {\n\n\t\tthis.touchCaptured = false;\n\n\t}\n\n\t/**\n\t * Convert pointer down to touch start.\n\t *\n\t * @param {object} event\n\t */\n\tonPointerDown( event ) {\n\n\t\tif( event.pointerType === event.MSPOINTER_TYPE_TOUCH || event.pointerType === \"touch\" ) {\n\t\t\tevent.touches = [{ clientX: event.clientX, clientY: event.clientY }];\n\t\t\tthis.onTouchStart( event );\n\t\t}\n\n\t}\n\n\t/**\n\t * Convert pointer move to touch move.\n\t *\n\t * @param {object} event\n\t */\n\tonPointerMove( event ) {\n\n\t\tif( event.pointerType === event.MSPOINTER_TYPE_TOUCH || event.pointerType === \"touch\" ) {\n\t\t\tevent.touches = [{ clientX: event.clientX, clientY: event.clientY }];\n\t\t\tthis.onTouchMove( event );\n\t\t}\n\n\t}\n\n\t/**\n\t * Convert pointer up to touch end.\n\t *\n\t * @param {object} event\n\t */\n\tonPointerUp( event ) {\n\n\t\tif( event.pointerType === event.MSPOINTER_TYPE_TOUCH || event.pointerType === \"touch\" ) {\n\t\t\tevent.touches = [{ clientX: event.clientX, clientY: event.clientY }];\n\t\t\tthis.onTouchEnd( event );\n\t\t}\n\n\t}\n\n}","import { closest } from '../utils/util.js'\n\n/**\n * Manages focus when a presentation is embedded. This\n * helps us only capture keyboard from the presentation\n * a user is currently interacting with in a page where\n * multiple presentations are embedded.\n */\n\nconst STATE_FOCUS = 'focus';\nconst STATE_BLUR = 'blur';\n\nexport default class Focus {\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t\tthis.onRevealPointerDown = this.onRevealPointerDown.bind( this );\n\t\tthis.onDocumentPointerDown = this.onDocumentPointerDown.bind( this );\n\n\t}\n\n\t/**\n\t * Called when the reveal.js config is updated.\n\t */\n\tconfigure( config, oldConfig ) {\n\n\t\tif( config.embedded ) {\n\t\t\tthis.blur();\n\t\t}\n\t\telse {\n\t\t\tthis.focus();\n\t\t\tthis.unbind();\n\t\t}\n\n\t}\n\n\tbind() {\n\n\t\tif( this.Reveal.getConfig().embedded ) {\n\t\t\tthis.Reveal.getRevealElement().addEventListener( 'pointerdown', this.onRevealPointerDown, false );\n\t\t}\n\n\t}\n\n\tunbind() {\n\n\t\tthis.Reveal.getRevealElement().removeEventListener( 'pointerdown', this.onRevealPointerDown, false );\n\t\tdocument.removeEventListener( 'pointerdown', this.onDocumentPointerDown, false );\n\n\t}\n\n\tfocus() {\n\n\t\tif( this.state !== STATE_FOCUS ) {\n\t\t\tthis.Reveal.getRevealElement().classList.add( 'focused' );\n\t\t\tdocument.addEventListener( 'pointerdown', this.onDocumentPointerDown, false );\n\t\t}\n\n\t\tthis.state = STATE_FOCUS;\n\n\t}\n\n\tblur() {\n\n\t\tif( this.state !== STATE_BLUR ) {\n\t\t\tthis.Reveal.getRevealElement().classList.remove( 'focused' );\n\t\t\tdocument.removeEventListener( 'pointerdown', this.onDocumentPointerDown, false );\n\t\t}\n\n\t\tthis.state = STATE_BLUR;\n\n\t}\n\n\tisFocused() {\n\n\t\treturn this.state === STATE_FOCUS;\n\n\t}\n\n\tdestroy() {\n\n\t\tthis.Reveal.getRevealElement().classList.remove( 'focused' );\n\n\t}\n\n\tonRevealPointerDown( event ) {\n\n\t\tthis.focus();\n\n\t}\n\n\tonDocumentPointerDown( event ) {\n\n\t\tlet revealElement = closest( event.target, '.reveal' );\n\t\tif( !revealElement || revealElement !== this.Reveal.getRevealElement() ) {\n\t\t\tthis.blur();\n\t\t}\n\n\t}\n\n}","/**\n * Handles the showing and \n */\nexport default class Notes {\n\n\tconstructor( Reveal ) {\n\n\t\tthis.Reveal = Reveal;\n\n\t}\n\n\trender() {\n\n\t\tthis.element = document.createElement( 'div' );\n\t\tthis.element.className = 'speaker-notes';\n\t\tthis.element.setAttribute( 'data-prevent-swipe', '' );\n\t\tthis.element.setAttribute( 'tabindex', '0' );\n\t\tthis.Reveal.getRevealElement().appendChild( this.element );\n\n\t}\n\n\t/**\n\t * Called when the reveal.js config is updated.\n\t */\n\tconfigure( config, oldConfig ) {\n\n\t\tif( config.showNotes ) {\n\t\t\tthis.element.setAttribute( 'data-layout', typeof config.showNotes === 'string' ? config.showNotes : 'inline' );\n\t\t}\n\n\t}\n\n\t/**\n\t * Pick up notes from the current slide and display them\n\t * to the viewer.\n\t *\n\t * @see {@link config.showNotes}\n\t */\n\tupdate() {\n\n\t\tif( this.Reveal.getConfig().showNotes && this.element && this.Reveal.getCurrentSlide() && !this.Reveal.print.isPrintingPDF() ) {\n\n\t\t\tthis.element.innerHTML = this.getSlideNotes() || 'No notes on this slide.';\n\n\t\t}\n\n\t}\n\n\t/**\n\t * Updates the visibility of the speaker notes sidebar that\n\t * is used to share annotated slides. The notes sidebar is\n\t * only visible if showNotes is true and there are notes on\n\t * one or more slides in the deck.\n\t */\n\tupdateVisibility() {\n\n\t\tif( this.Reveal.getConfig().showNotes && this.hasNotes() && !this.Reveal.print.isPrintingPDF() ) {\n\t\t\tthis.Reveal.getRevealElement().classList.add( 'show-notes' );\n\t\t}\n\t\telse {\n\t\t\tthis.Reveal.getRevealElement().classList.remove( 'show-notes' );\n\t\t}\n\n\t}\n\n\t/**\n\t * Checks if there are speaker notes for ANY slide in the\n\t * presentation.\n\t */\n\thasNotes() {\n\n\t\treturn this.Reveal.getSlidesElement().querySelectorAll( '[data-notes], aside.notes' ).length > 0;\n\n\t}\n\n\t/**\n\t * Checks if this presentation is running inside of the\n\t * speaker notes window.\n\t *\n\t * @return {boolean}\n\t */\n\tisSpeakerNotesWindow() {\n\n\t\treturn !!window.location.search.match( /receiver/gi );\n\n\t}\n\n\t/**\n\t * Retrieves the speaker notes from a slide. Notes can be\n\t * defined in two ways:\n\t * 1. As a data-notes attribute on the slide \n\t * 2. As an
