Clean preprocessing with held out test set

[jjm82]

• Choose best data set from Kaggle
• create new file preprocessing.py
• put dataset into this file and hold out a randomly sampled test set which is balanced
• i.e. output a train/validate set and a held-out test set
• upload these files to the folder for permanent use (training/validate for CV and hyper-parameter tuning, test for final evaluation)