Added min_loss

[marnunez]

From base

[aclouthier]

Have you seen that stopping criteria used previously? I've seen recommendations to base the stopping criteria on validation loss as opposed to training loss. I find the loss is pretty noisy, so setting a hard limit like that might stop the training a bit early.

[marnunez]

Truth be told, since I found your repository, I'm devouring a pytorch course :). So I haven't seen much of anything previously. I'll look into validation loss, see what I can modify