covariance matrix with w or without w

[ScorpioBao]

Hello, author, in the cov method of loss_reweighting.py file in your code, you define the operation of the covariance matrix with and without w, but it seems that the calculation of w does not match the definition in your paper, I do not know whether it is a code error or I understand the error.

[xxgege]

Hi, actually, the input w is after a softmax function, thus the sum of w is already 1.
Please see line 23 of training/reweighting.py, where the function is called.