Hand Written Digit Recognition using Multi-Layer Neural Network

Introduction

Hand written digit recognition using multi-layer neural network build using only python's numpy library. The included pre-trained model achieved an accuracy of 95.54 % on MNIST dataset.

Requirements

• numpy
• mnist

Usage

• train a model

  python recognizer.py --train model_name

• Continue training from a specific model

  python recognizer.py --train model_name --loadmodel old_model_name

• Evaluate pre-trained model

  python recognizer.py --evaluate trained_weights.npz