Neural Network from Scratch (aka SigmoidNN)

This project implements a simple yet powerful neural network using scary mathematical operations, such as the sigmoid activation function, cross-entropy loss, and matrix operations😖. The network has been tested with the MNIST dataset of handwritten digits and achieved an 97.99% accuracy with the chosen parameters.

You can customize the layers, activation functions, and data to train and test the network for your specific needs.

---

Checkout our python library sigmoidNN

python3 -m pip install sigmoidNN

Features

• Fully Customizable Neural Network: Define your own architecture, including the number of layers and neurons.
• Basic Mathematical Foundations: Built using fundamental concepts, including the sigmoid activation function and cross-entropy cost function.
• Pre-trained Models: Explore trained instances stored in the src/trained_instances folder.
• High Accuracy: The network achieves up to 97.99% accuracy on the MNIST dataset with the default configuration.
• Modular Code: Easy-to-understand and modify for experimentation with different datasets and configurations.

---

Folder Structure

project-root/
|
├── src/                # Source code for the neural network
│   ├── network.py      # Core neural network implementation
│   ├── utils.py        # Helper functions for data processing
│   ├── train.py        # Training script
│   └── test.py         # Testing script
|
├── data/               # Folder for datasets (e.g., MNIST)
│   └── mnist/          # MNIST dataset files
|
├── src/trained_instances/   # Pre-trained model instances
│   └── net[<accuracy>].json # Example trained model for MNIST
|
└── README.md           # Project documentation (this file)

---

Getting Started

Prerequisites

Make sure you have the following installed:

• Python 3.8+
• NumPy
• dataset (you can find one in /data folder)

Installation

1. Clone the repository:

   git clone https://github.com/Anas-github-acc/SigmoidNN
   cd SigmoidNN

2. Create the environment (it's recommended to use a virtual environment)::

   python3 -m venv venv
   source venv/bin/activate   # For MacOS/Linux
   venv\Scripts\activate      # For Windows

3. Install dependencies:

   pip install -r requirements.txt

Qucik Start

Run on the Colab

• download the mnist.data.gz and upload in the colab file
• or upload your own data, edit the path in cell 3 and adjust the parameters like layers
• run all the colab

Run the network with the MNIST dataset:

python src/main.py

This will save the trained model in the src/trained_instances/ folder ( their you will find pre-trained result's instance with 96-98% accuracy with chosen parameters You can load these instance and use

Load the instance

Loading the trained instance in src/trained_instances/:

from network import load
load(/Path /to /Instance)

Customizing the Network

To customize the network, modify the parameters in the network.py file, such as:

• Number of layers & Neurons per layer (this take input in list)
• Learning rate
• Regulation
• Batch size
• Activation function

---

Results

The network achieved the following results on the MNIST dataset:

• Accuracy: 97.99%
• Training Parameters:
  • Activation function: CrossEntropyCost
  • Regularization: 0.7
  • Learning Rate: 0.5
  • Batch Size: 10
  • Number of Epochs: 15
  • Architecture: [784, 100, 10] (Input Layer -> 1 Hidden Layers -> Output Layer)

Pre-trained results are available in src/trained_instances/net[<accuracy>].json. You can load and test this model for verification or use.

---

Contributing

Contributions are welcome! Feel free to fork the repository, make improvements, and submit a pull request.

Acknowledgments

I highly recommend the book by Michael Nielsen introducing neural networks and deep learning: https://goo.gl/Zmczdy