MNIST Neural Network from Scratch

This script is a basic implementation of a Neural Network in Python using NumPy from scratch. It's designed to classify handwritten digits from the MNIST dataset.

Features

• Loads the MNIST dataset from local files.
• Preprocesses the data (normalizes pixel values and one-hot encodes labels).
• Trains a neural network on the dataset using mini-batching, sigmoid activation and stochastic gradient descent.
• Evaluates the trained model's performance on a test set.
• Saves a plot of the model's accuracy over training epochs.

Requirements

• Python 3
• NumPy
• Matplotlib

Usage

1. Update labels_file_path and images_file_path with your local paths to the MNIST dataset.
2. Run the script with Python: python mnist_cnn.py.

Outputs

• Console output of model accuracy every 10 epochs and final test accuracy.
• A 'accuracy.png' plot file showing model accuracy over epochs.