Add support to record selection information for a datarun

[myaooo]

I am working on ATM-VIS, a web-based GUI for atm.
Currently, the db only stores classifier information. Users have no clues of how each hyperpartition is chosen at each step, and they have no hints on which selector to use. Providing the selector information at each step might help the users better understand or trust the ATM procedures (or leads to improvements of the algorithm).

There is a need to provide selection information (rewards/arm scores) during (or after) a data run to the users (so that the users can choose a selector that suits their needs).

For example, we would like to plot how the rewards of each hyperpartition (or method) changes as the datarun proceeds

[myaooo]

It seems that we would need to modify the code in BTB to make this happen?

I checked the BTB code. It seems that the Selector does not seem to provide an api to retrieve the arm scores (one score for each hyperpartition) for a given step (or given score histories)?

[micahjsmith]

What about the compute_rewards API method? https://hdi-project.github.io/BTB/btb.selection.html#btb.selection.selector.Selector.compute_rewards

The ATM code would have to be modified to get the rewards from BTB and log them accordingly at each step.