I was running Text Classification on my main desktop machine, just to test. It took a bit longer than expected, and the computer crashed after 3 weeks.

• It would be great to see how quickly the training is progressing. User can cancel if it seems too slow.
• During training it can also encourage user to complete the training. Currently, I did not know if it was 10% or 95% complete.

I think, just number of epochs completed and remaining to train would be enough. Estimated time to completion is probably not needed.

Perhaps the number of max epochs could also be user adjustable in some future version. Likewise, if it could autosave checkpoint and knew how to continue next time (or even guide for advanced cases to recover the file) would be nice.