Early stop training setup

Oct 9, 2015 at 6:39 AM
Is there any configuration about early stopping training before reach the max epoch?

For example, when the performance on validation set does not improve in given epochs, the training stops.
Oct 9, 2015 at 6:44 AM
Yes, search AutoAdjust in CNTK book.
Oct 9, 2015 at 7:27 AM

I've read the description of AutoAdjust in Page 62 and the sample in Page 162.

AutoAdjust is used to adjust learning rate during training, but what is the minimum learning rate below which the training stops?
It seems CNTK book does not have parameter to set this minimum learning rate?

BTW, is it possible to early stop training if I just use constant learning rate?