Goal: Create a model that recognizes images of numbers 0-9.
MNIST1 is first used to create 540,000 extra images with elastic distortion.
MNIST2 is a pure numpy implementation of a 2 layer neural network, the model trained had an accuracy of 96.9% accuracy on test data.
MNIST3 uses tensorflow.keras and implements a CNN, accuracy of 99.59%. (saved in the h5 file)