This is a gui for training and testing a digit detection system based on the neural network described in Tariq Rahid's book "Make your on neural network".
The gui relies on kivy. The neural network makes use of scipy, numpy and matplotlib. After installing the dependencies just run neural_draw.py.
This repository includes the gui and the neural network. It does not include the mnint test data however. You can get it here and copy it to mnist_dataset/mnist_train.csv. If you check the Save in test data box, the result of your test will be added to mnist_dataset/my_test_data.csv. This data can later be used to improve your test data.
The weights for the network are provided in the weights_who.npy and weights_wih.npy files so the network does not need to be trained everytime you start the gui. Delete these files if you want to use a new set of training data.
- Make the training data file location editable in the gui
- Improve the process of saving, cropping and analyzing the drawing area
- Create a setup.py file