This repository contains a neural network library built from scratch using only Python and NumPy. It aims to provide a fundamental understanding of neural network mechanics without relying on high-level deep learning frameworks.

• Pure Python and NumPy Implementation
  Designed for clarity and educational purposes, demonstrating how neural networks work at a low level.

• Modular Design
  Separate modules for key components of a neural network.

• Activation Functions
  Includes common activation functions used in neural networks.

• Loss Functions
  Provides various loss functions for different types of machine learning tasks.

• Metrics
  Offers metrics to evaluate model performance.

• Neural Layers
  Implements different types of neural layers, allowing for flexible network architectures.

• Neural Network
  The main class for building, training, and evaluating neural networks.

• Binary Classification
  An example script (MainBinaryClassification.py) to demonstrate binary classification tasks.

• Multiclass Classification
  An example script (MainMulticlassClassification.py) to showcase multiclass classification tasks.

• Utility Functions
  A Utils.py module for helper functions.

• Python 3.12 or higher
• NumPy

No specific installation is required. Simply clone the repository:

git clone https://github.com/denizbilgin/NeuralNetworkLibrary.git
cd NeuralNetworkLibrary

You can easily define your own neural network model and train it using the provided classes:

# Determine layers
hidden1 = NeuralLayer(20, ReLU())
hidden2 = NeuralLayer(7, ReLU())
hidden3 = NeuralLayer(5, ReLU())
output_layer = NeuralLayer(1, Sigmoid())

# Create neural network
my_network = NeuralNetwork(
    train_x, train_y,
    [hidden1, hidden2, hidden3, output_layer],
    BinaryCrossEntropy(),
    [accuracy],
    0.001
)

# Train the network
costs = my_network.fit(2000)

# Make predictions with the trained model
predictions = my_network.predict(test_x)

# Evaluate the test set
my_network.evaluate(test_x, test_y)

You are free to configure different architectures, activation functions, and loss metrics according to your task.

Contributions are welcome! If you have suggestions for improvements, new features, or bug fixes, please open an issue or submit a pull request.

For any questions or feedback, feel free to reach out.