WIP per-class metrics

[swsuggs]

For #492

[swsuggs]

@davidslater The poisoning scenario adds this metric automatically, right in the code. Is this an acceptable way to make it available to other scenarios through the config/metric/task section?

[davidslater]

I like adding it to the config part of instrumentation.

[swsuggs]

@davidslater So far I've added (beside per-class accuracy) a confusion matrix, and precision and recall for each class. Two questions: Will you check that I'm computing the right thing (I described my understanding in the comments), and that the output is in a sufficiently useful format (dict, array, etc)? Can you think of any other metrics or statistics that would be nice to have?

[davidslater]

See my two recent comments. Beyond that, I think what needs to be done is: A) remove the changes to armory/instrument/config.py, and armory/utils/config_schema.json. B) move the new functions from statistical to task, since we're registering them as task populationwise metrics anyhow. C) Probably easiest to rebase off of develop.