Python for Deep Learning - Homework Repository

Welcome to the Python for Deep Learning (DL) course homework repository!
This repository is organized week-by-week to showcase projects and hands-on exercises developed throughout the course.

---

📅 Weekly Homework Overview

📘 Week 01–02: Maze Game

A Python-based interactive Maze Game built with Tkinter.

• Procedural maze generation
• Manual or auto-solve navigation
• GUI with arrow-key support
• 📄 See: week01-02/README.md

---

🏠 Week 03: Real Estate Price Prediction

A linear regression model using PyTorch to predict housing prices based on features such as property age, area, and distance to the city center.

• Trained using stochastic gradient descent
• Includes training curve and MAPE evaluation
• 📄 See: week03/README.md

---

🧠 Week 04: Perceptron Implementation

Implementation of a simple perceptron using:

• scikit-learn for baseline experimentation
• PyTorch for custom hands-on model
• Comparison of normalization, loss curves, and confusion matrices
• 📄 See: week04/README.md

---

🌀 Week 06: CNN Improvements Across Datasets

Enhanced a convolutional neural network (CNN) and applied it to MNIST, FashionMNIST, and EMNIST datasets.

• Improved architecture with multiple convolutional layers, batch normalization, and dropout
• Training/validation metrics and misclassified samples visualized using TensorBoard
• 📄 See: week06/README.md

---

⚡ Week 08: Lightweight CNN Model

Developed a lightweight CNN architecture and applied it to MNIST, FashionMNIST, KMNIST, and EMNIST.

• Utilized depthwise separable convolutions and Squeeze-and-Excitation blocks
• Incorporated data augmentation, learning rate scheduling, and early stopping
• Configured num_workers for efficient data loading during GPU training
• Visualized model performance and misclassifications via TensorBoard
• 📄 See: week08/README.md

---

🐶 Week 09: Dog Breed Classification

Applied transfer learning to classify 12 dog breeds using GoogLeNet, ResNet-18, EfficientNet-B0, and Swin Transformer.

• Dataset collected via Bing Search
• Aggressive data augmentation and dropout regularization
• TensorBoard logging and confusion matrices for evaluation
• 📄 See: week09/README.md

---

📊 Week 10: Crowd Counting

Built a crowd counting regression model using a fine-tuned pre-trained CNN.

• Dataset: Aerial crowd images from ShanghaiTechDataset
• Fine-tuned with Smooth L1 Loss (Huber Loss) and evaluated with custom Model Score
• Emphasis on balancing accuracy, model size, and efficiency
• 📄 See: week10/README.md

---

🚀 How to Use

1. Clone the repository:

git clone <your-repo-url>
cd python-for-dl-homework

2. Navigate to a specific week's folder and follow its instructions:

cd week03
python price_prediction.py

---

📜 License

All content in this repository is licensed under the MIT License.
You are welcome to explore, learn from, and build upon this work.