trained Weight not applied to Sample process of ScreenerNet ?

[svjack]

In the example of ScreenerNet, the resample process of train process seemed should apply with
trained weight of sample, But it seemed like the train loader simply retrieve trained samples in ordinarily
pytorch dataloader manner in the file sent.py in ScreenerNet dir ?
So the only effect of ScreenerNet is the grads update of Main NetWork ?
And I have not see the PrioritizedExperience Replay(PER) Process in the code (adjust the weight for sampling)

I am Confused with this, it may related with my misunderstanding, Please give me an explaination.
@TobeyQin

[waitwaitforget]

Hi, there,
I'm also interested in ScreenerNet, maybe we can discuss it together. Actually, I've tried to use ResNet to replace the basenet here, and the results are not satisfying, it seems that screenernet does not give a better convergence or much slower than the baseline.