I was reading the code by Aakash Nain on AugMix https://github.com/AakashKumarNain/AugMix_TF2/blob/master/utils.py and found the word 'patience' under the Class CTEarlyStopping. Not the first time I have encountered this word but I'm a noob in these terminologies. Looking for a quick explanation.
When you train something, it starts from high loss and during the training, the loss goes down and downer (accuracy gets higher).
After a limited number of epochs, the loss which is low enough, won't go lower (accuracy doesn't get higher). In this situation, it seems that training more epochs is useless because your train is almost complete.
Now in this moment, the patience value tells the trainer, how many epochs it must continue after the loss stopped from decreasing. If you don't set patience value, the training will continue for all the epochs you set, even if your training results are not getting any better. This value saves you some time and energy (by resting the gpu).