training with custom data

[eunkoh]

Hi team,

I am currently working with drubinstein/wip-training branches with my custom data. Under the branch, model.py, I can see multiple losses and smoothing parameters I can control, and I wonder if I can get some information from previous training (e.g., previous tensorboard logs etc). For example, how loss worked, some information "contour_loss", "onset_loss", and "note_loss". If I can get that info, that would be greatly helpful for understanding my current training better. Any suggestions?