This project is a Python implementation of a feed-forward neural network originally written in C. It demonstrates:

• A simple neural network architecture with one hidden layer.
• Backpropagation training with momentum.
• Xavier initialization of weights.
• Checkpointing: saving and loading network parameters from a binary file.
• Custom logic for handling logical operations (e.g., XOR, XNOR, OR, AND, NOR, NAND).

The code includes:

• A linear congruential generator for reproducible randomness.
• Logical operation functions to generate target outputs.
• Activation functions: the sigmoid and its derivative.
• A Network class that defines the neural network structure and performs feed-forward predictions.
• A Trainer class that implements backpropagation with momentum to update the network parameters.
• Utility functions for printing, saving, and loading the network parameters.

The network is trained using two phases and periodically saves checkpoints to allow resuming training from a saved state.

• Feed-Forward Neural Network: Configurable number of inputs, hidden neurons, and outputs.
• Backpropagation with Momentum: Improves training efficiency.
• Xavier Initialization: Helps speed up convergence during training.
• Checkpointing: Automatically saves network parameters, allowing training to be resumed.
• Logical Operation Tasks: Demonstrates training on logical functions like XOR, XNOR, OR, AND, NOR, and NAND.

• Python 3.x
• NumPy

1. Clone the repository:

   git clone https://github.com/firstpixel/neural-network.git
   cd neural-network

Install dependencies:

pip install numpy

Usage Run the main script to start training the neural network:

python main.py

Initialize the network with 2 inputs, 10 hidden neurons, and 6 outputs. Attempt to load a checkpoint (if available). Train the network in two phases, saving checkpoints at regular intervals. Print initial and final results, along with the final network parameters. Code Structure Rand() Function: Implements a linear congruential generator for reproducible randomness.

Includes xor_op, xnor_op, or_op, and_op, nor_op, and nand_op.

sigmoid and sigmoid_prim (its derivative).

Sets up the neural network, performs Xavier initialization, and handles feed-forward prediction.

Computes gradients and updates weights using backpropagation with momentum.

print_network(): Prints current network parameters. network_save(): Saves network parameters to a binary file using NumPy. network_load(): Loads network parameters from a binary file. main() Function: Coordinates the training process, manages checkpointing, and prints results.

Contributions are welcome! Feel free to fork the repository and submit pull requests with improvements or additional features.