The data for this competition were taken from the MNIST dataset. The MNIST ("Modified National Institute of Standards and Technology") dataset is a classic within the Machine Learning community that has been extensively studied. More detail about the dataset, including Machine Learning algorithms that have been tried on it and their levels of success, can be found at http://yann.lecun.com/exdb/mnist/index.html.

This competition is evaluated on the categorization accuracy of your predictions (the percentage of images you get correct).

A simple learning experience creating:

1. A single layer neural network (single_layer.py)
2. A multi layer neural network (multi_layer.py)
3. A convolutional neural network (convolutional.py)

The best performing will be submitted to kaggle.

The best performing algorithm was, unsurprisingly, the convolutional neural network. Submitted to Kaggle, it had an accuracy score of .99571 for the first submission which placed it at 72 place out of 1414 participants (this placement changes as more people submit projects).

I believe that I could achieve even higher accuracy but I don't think it is that important. I don't want to overfit the leaderboard. I believe that my convolutional neural net is performing well and does not need to be tweaking to gain better results.

Special Thanks to Hvass-Labs for a great into to TF NN. His guidance helped in developing these.