A question about your loss_coteaching function

[nihaomiao]

Hello, I find that in your class "loss_coteaching", the parameters you passed are y_1, y_2, which are already the log_softmax results. But you then used the cross_entropy, which has combined the log_softmax and nll_loss. This will use log_softmax twice. I am not sure whether I am wrong. Or do this problem not occur in your pytorch version?
By the way, even though using log_softmax function twice, your code is still right~

[bbdamodaran]

I also observed the same issue, but I tested using log_softmax only once. I did not observe the difference in the results. Please let me know if you observe different results. One more issue is that in return function of loss, the loss value is normalized twice

[nihaomiao]

OK~

[guixianjin]

I don't observe that y_1, y_2 are already the log_softmax results, have the code has been changed and fixed? @nihaomiao @bbdamodaran

[henrypickler]

This was fixed in 4d32d1c. Note that even if the results were already log_softmax transformed this would not change the results since applying log_softmax twice is the same as applying it only once. It also does not change the ordering for selecting samples for each model meaning it didn't had any effect other than a bit of extra computation.