Perceptron and Two-Layer Neural Network from Scratch

This project implements a single-layer Perceptron and a two-layer neural network (MLP) using only NumPy, without using any high-level machine learning libraries (like TensorFlow or PyTorch).
It serves as an educational demonstration of how neural networks work under the hood.

---

📌 Objectives

• Understand and implement the basic components of a neural network:
  • Initialization of weights and biases
  • Activation functions
  • Cost/loss computation
  • Gradient descent (manual backpropagation)
  • Training loop

---

🧠 Implemented Models

1. Perceptron (single layer)
   • Binary classification
   • Linear activation
2. Two-layer neural network
   • Hidden layer with non-linear activation (ReLU or sigmoid)
   • Output layer with softmax or sigmoid (depending on the task)
   • Manual gradient calculation and training

---

🛠️ Technologies

• Python 3.x
• NumPy
• Jupyter Notebook

---

🧪 How to Run

1. Clone the repository:

   git clone https://github.com/your-username/perceptron-from-scratch.git
   cd perceptron-from-scratch
   

2. Open the notebook:

   jupyter notebook perceptron.ipynb
   

3. Follow the cells and run them step by step.

---

📝 File Structure

perceptron-lab/ ├── README.md
├── requirements.txt
├── notebook/ │ └── perceptron.ipynb
├── src/
│ ├── perceptron.py
│ └── mlp.py
├── data/
│ ├── testset.hdf5
│ └── trainset.hdf5
|
├── images/
│ └── └── .gitignore

📊 Dataset

The code includes toy data generated with NumPy (e.g. XOR, linearly separable datasets).

You can easily plug in any 2D dataset like sklearn.datasets.make_classification or make_moons.

🎓 Author

Kodjo Jean DEGBEVI Student in Artificial Intelligence & Big Data