Handwritten Digit Recognition

Neural Network

This is a feed-forward neural network based only in Numpy, which implements and trains a 4 layer neural network. The neural network defaults to 4 layers, but you are able to train an N layer neural network. The MNIST database is used for both training and testing.

Usage

pip3 install -r requirements.txt

Install the dependencies from requirements.txt

python3 paint.py

This will load model that is saved in /model and allow user to draw digits on UI.

Train

python3 train.py

This will train and save model to /model. By default this will train a neural network with an input layer of 786 nodes, two hidden layers of 512 and 256 nodes respectfully, and one output layer of 10 nodes for 10 epochs. You can adjust the hidden layers and epochs as you see fit.

Now you can start playing around with the code!