Provide a recepie for training a model on MovieLens data (20M and 1M).

This should include the following:

1. Data converter. Consider fixing/adjusting this script

• Train/Eval/Test split. Training test ratings should come before any Eval and Test ratings. See section 3.1 (Experiment setup) of this paper for details and Netflix converter for example.

2. An example train.sh, and test.sh for 20M and 1M MovieLens data sets which will train/eval and test the model correspondingly.

3. Choose model architecture and hyper parameters that give best eval RMSE you can get.