This multi layer perceptron network using just numpy is pretty much entirely stolen from the internet, because lazy.
The train.py file contains an aptly named function, load_data2(), to conduct extraction and rough normalization of the input data, which has received some ETL (all integer values had commas stripped out, headers removed).
Create a python virtual environment using python 2.7 and the requirements file using conda or virtualenv.
To simply see a single training and test of a single neural network, run: python train.py
To see some cool graphs playing around with different neural network hyperparameters, run: python train_cool.py
**Note for conda users, you may need to run conda install python.app and then execute train_cool.py with pythonw