Largely a maintenance, refactoring, and bug-fix release. Call it housekeeping.
Refactors the model class for improved post-hoc interrogation and re-use of model sub-components. This resulted in a big diff, but conceptually little has changed.
Refactors losses to account for new model structure, adds some regularizations
Adds the ability to train multiple unique domain adversaries, closing long-standing Issue #7
Adds utilities to extract model training metrics from Tensorboard event logs in utils
Fixes several bugs that were preventing tests from passing. For future reference, it looks like the Travis CI VMs have pretty limited memory, so we need to be judicious about which of the actual model training tests we run with every CI push.
Largely a maintenance, refactoring, and bug-fix release. Call it housekeeping.
lossesto account for new model structure, adds some regularizationsutils