How to find the best model when training

[quocnhat]

trafficstars

Hi, Thanks very much for sharing your awesome works. Due to the variation of the dataset ( hard & easy examples in each iteration), the out loss is unstable at all. So how can we determine the best model when training?

[DerekChen88]

Same problem. I wonder how to calculate AP during training in order to find the best model. The provided evaluation method seems cannot be used in the training.

[quocnhat]

for now, The simplest method is to save the best loss checkpoint crosses the number of epochs. I will evaluate that method when finish.