`train_vae` should dump an image at each epoch, and `train_{vae,nn}` should run `cryodrgn analyze` at the end.

[adamlerer]

[Guillawme]

I agree it's a good default behavior to automatically analyze the last iteration.

But it would be nice if this could be turned off too. When using a cluster and a queuing system, one might want to request different resources for analysis than for training.

[michal-g]

fyi, we are working on a version of this behaviour (auto analysis of the final epoch at the end of training, with a flag for train-only) for an upcoming release!