Model training debugging capabilities

[akashshah59]

I was wondering if we could enable Model debugging and logging capabilites.

I've just been wanting to plot the loss curve for Generator and Discriminator, While I admire the design of how ydata has been built, it would be nice to have some tensor-board like plots epoch curves automatically flushed out, I've also been thinking of building visualizations do see how distribution shifts occur for Generator over time. I perused through base.py however, all I can see is training_params, which does have a log_frequency , but maybe we can add a /logging_dir as well.

I'd be willing to work on this, let me know what you guys think?