This program is a shallow Neural Network that is trained to recognize digits written in a 5x3 box
The architecture is a fully connected shallow Neural Network with only 1 hidden layer.
With a training size of 27 the model used a learning rate of 0.3 and 1,000 iterations. The model was able to get a 100% accuracy on 25 validation examples. Both the forward propogation and backward propogation algorithms were made by me using only mathematical formulas as a reference. The specific activation function was I used was a sigmoid function and no Cost function was used. Additionally, since the problem is relatively simple, biases were not used and only weights were.
This project is made completely from scratch and the only library used was to read the training and validation data from .csv files.