MNIST Digit Recognizer

This project is a simple neural network built from scratch to recognize handwritten digits from the MNIST dataset. The model is implemented using only NumPy and focuses on understanding the fundamentals of neural networks.

Project Structure

• data/: Contains the dataset used for training and testing.
• notebooks/: Jupyter notebook with the main implementation and experiments.
• scripts/: Python scripts for training and utility functions.
• models/: Directory to save trained model parameters.
• results/: Contains results such as plots of training performance.

Installation

1. Clone the repository:

   git clone https://github.com/Rikhil-Nell/MNIST_digit_reader.git
   cd MNIST_digit_reader

2. Set up a virtual environment and install dependencies:

   python -m venv venv
   source venv/bin/activate  # On Windows use `venv\Scripts\activate`
   pip install -r requirements.txt

Usage

1. Open the Jupyter notebook:

   jupyter notebook notebooks/mnist-digit-reader.ipynb

2. Run the cells to train the model and visualize the results.

Results

Here is an example of the accuracy and loss curves over training iterations:

Future Work

• Implement additional layers and experiment with different architectures.
• Add more sophisticated data preprocessing and augmentation techniques.
• Explore hyperparameter tuning to improve model performance.