Kaggle digit recognizer

This script can be used to predict Digits based on hand-written images. It uses training and testdata provided by kaggle. It will have multible implementations using different algorithms.
training set size: 42000 test set size: 28000

Current implementations

Neural network - no framework

lambda: 1
iterations: 2000
hidden layers: 2 (including output layer)
hidden layer size = 200
training set accuracy: 98.259524%
test set accuracy: 0.9451%

Neural network - neural network toolbox

iterations: ~ 100 hidden layer sizes: 50 30 20 test set accuracy: 97,657%