General-Purpose Neural Network Framework

This project provides a reusable neural network framework implemented for Exercise 2 in the "Computational Models of Learning" course.
The framework supports arbitrary input/output sizes, multiple layers, any activation function, and both scalar and vector-valued outputs.

---

📁 Project Structure

backpropagation_project/           # Project root
├── neuralnet/                     # General-purpose neural network framework
│   ├── __init__.py
│   ├── layers.py
│   ├── network.py
│   ├── training.py
│   ├── utilities.py
│   ├── visualization.py
│   └── assertions/
│       ├── __init__.py
│       ├── layer_assertion.py
│       ├── network_assertion.py
│       └── training_assertion.py
├── scripts/
│   ├── part1.py
│   └── part2.py
├── .gitignore
├── LICENSE                         # MIT License
├── README.md   
├── requirements.txt                # Python dependencies
└── setup.cfg

---

📚 Features

• Supports arbitrary input and output dimensions
• Custom activation functions per layer
• Layer-wise modular assertions
• Visualizations for model behavior
• Separation of general-purpose code and exercise logic

---

🛠 Setup

1. Create a virtual environment

Linux / macOS:

python3 -m venv .venv
source .venv/bin/activate

Windows:

python -m venv .venv
.venv\Scripts\activate

2. Install dependencies

pip install -r requirements.txt

---

▶ How to Run

Always run from the project root (backpropagation_project/) using module mode:

python -m scripts.part1
python -m scripts.part2

This ensures all relative imports and packages are resolved correctly.

---

📝 Important Notes

• Python ≥ 3.4
• numpy
• matplotlib

📄 License

MIT License. See LICENSE for details.

👤 Author

• Name: Or Fadida
• Email: orfadida@mail.tau.ac.il
• GitHub: orfadida2000