create a database for tracking runs

[niallmcc]

from @cjmerchant

Set up a database to keep a record of the train/test dataset of each experiment/run, the output data, and the quality of the results (measured by RMSE, MAE, Mean Pattern Correlation, among others to be added).

think about including random seeds, but it remains to be seen if we can truly have reproducible results

@niallmcc will do the first sprints work, using SQLlite.

[niallmcc]

Things we need to track:

• date and time
• random seed?
• input data file names (train and test) and possibly hashes
• input model (if retraining)
• hyper-parameters
• evaluation metrics
• output model name

Some fields may be stored as JSON encoded strings so we can easily extend them.

[niallmcc]

We some prettier output for dumping the database using query_database.py