One should be able to stop training if a desired criterium is satisfied, e.g., if training loss is below a prescribed threshold.
At the moment, we only support training for a fixed number of epochs.
Let us introduce a new parameter stopping_criterium that takes the current training status (like epoch, loss/train, etc.) and returns True/False if training should be stopped. Per default, we can set this criterion to loss/train < 1e-5.
One should be able to stop training if a desired criterium is satisfied, e.g., if training loss is below a prescribed threshold. At the moment, we only support training for a fixed number of epochs.
Let us introduce a new parameter
stopping_criterium
that takes the current training status (like epoch, loss/train, etc.) and returns True/False if training should be stopped. Per default, we can set this criterion toloss/train < 1e-5
.