/Neural_Network_with_Layers

Primary LanguagePython

Neural Network with Layers

This repository contains a simple implementation of a neural network from scratch using only the NumPy library. The model was designed to provide a clear understanding of how a neural network operates internally, without relying on high-level frameworks such as TensorFlow or PyTorch.

Overview

This project demonstrates the construction of a fully-connected, multi-layer neural network. The primary goal is to implement and train a neural network from scratch, understand the forward and backward propagation processes, and optimize the network using gradient descent.

Features

  • Custom Neural Network Architecture: Implemented using only NumPy.
  • Forward and Backward Propagation: Manually calculated gradients.
  • Gradient Descent Optimization: Hand-coded to adjust the weights and biases of the network.
  • Configurable Hyperparameters: Easily modify the number of layers, neurons, learning rate, and activation functions.
  • Training and Evaluation: Designed to handle a simple dataset for training and validation purposes.

Repository Structure

  • dataset/ : Contains the dataset used for training the neural network.
  • images/ : Includes visualizations and plots related to the model's performance.
  • .gitignore : Specifies files and directories to be ignored by Git.
  • Model.py : The core implementation of the neural network, including all functions for forward pass, backward pass, and optimization.

Getting Started

Prerequisites

Ensure you have Python installed along with NumPy. You can install the required packages using:

pip install numpy

Running the Model

To train and evaluate the model, simply run the Model.py script:

python Model.py

The script will:

  • Load the dataset.
  • Initialize the neural network with the specified parameters.
  • Train the model using gradient descent.
  • Evaluate the performance on a validation set.

Configuration

You can configure the neural network's hyperparameters directly within the Model.py script. Parameters such as learning rate, number of layers, and activation functions can be adjusted to experiment with different network setups.

Results

The trained model achieves an accuracy of over 80% on the provided dataset, demonstrating the effectiveness of a simple neural network even without complex frameworks.

Visualizations

Refer to the images/ directory for plots and visualizations of the training process, including loss curves and accuracy graphs.

Contributing

Contributions are welcome! If you have suggestions for improvements or additional features, feel free to open an issue or submit a pull request.