Sure! Here's the full content formatted in Markdown for direct use in a README.md file:

# AFFEC Dataset Processing and Training Instructions

This repository contains code for processing the **AFFEC** dataset and training models using the processed multimodal data. Follow the instructions below to prepare the data and run training.

---

## 📦 1. Download the Dataset

Download the following components from the AFFEC dataset hosted on Zenodo:

- **Eye Tracking Data**
- **Pupil Data**
- **Face Analysis Data**
- **Electrodermal Activity (EDA) and Physiological Sensors**
- **Self-Annotations**

You can download the dataset from the following link:

🔗 [https://zenodo.org/records/14794876](https://zenodo.org/records/14794876)

Once downloaded, extract the contents into a directory of your choice.

> ✅ Make sure the folder contains:
> - `participants.tsv`
> - Subfolders for each participant (e.g., `sub-xxx/`) with their respective data files

---

## 🧪 2. Generate the Pickle Dataset

The `pickle_generation.py` script processes the multimodal sensor data and merges it into a single pickle file (`dataset.pkl`), which can then be used for training.

Open your terminal and run:

```bash
python pickle_generation.py --dataset_path /path/to/dataset

Replace /path/to/dataset with the actual path where you extracted the dataset.

🗃️ After this step, a file named dataset.pkl will be created in the dataset directory.

---

🧠 3. Train the Model

Once the dataset is processed, you can run the training pipeline. Make sure the correct path to the pickle file is provided.

Run the following command:

python multiphase_simple.py --data_path dataset.pkl

📝 This script trains a model on the dataset and saves the evaluation results in:

• results.csv

---

✅ Summary

Step	Description
1️⃣	Download AFFEC dataset from Zenodo
2️⃣	Run pickle_generation.py to preprocess the data
3️⃣	Run multiphase_simple.py to train and evaluate the model

---