create an internal train/test split to assess different prediction approaches

[kaichop]

take 80% of the train.csv and create a new newtrain and take 20% and create a newtest file, so that we can evaluate different approaches on known results. This should be done in a random fashion rather than taking the first 80% lines.

[umahsn]

Completed.

[umahsn]

Located here: /mnt/isilon/wang_lab/shared/Belka/raw_data

[Euchiz]

I generated the parquet version of the splits to the same location.