Evaluation during training and it’s statistics

Open chenzhutian opened this issue 3 years ago • 2 comments

Is your feature request related to a problem? Please describe. I I have two scenes, one for training and the other for testing. During the training, I would like to test the model in the testing scene every 20k steps, and collect and plot the statistics on Tensorboard. So that I can see whether the model really learn something instead of overfitting the training scene. Is this possible? Any suggestions for doing this?

Describe the solution you'd like can easily switch between training and testing phrases by control some environment parameters.

Mar 07 '22 15:03 chenzhutian

Hi @chenzhutian,

This is a useful feature. I logged it internally under MLA-2465 and we will prioritize accordingly.

Mar 14 '22 17:03 maryamhonari

Hi @maryamhonari any pointers if I want to have a quick hack for this feature? (working on a paper submission....

Apr 11 '22 05:04 chenzhutian