incorrect lr adjust

[Zyun-Y]

hi,

I found an error in your code.

in the training part, you wrote 'if i + epo * len(dataloader) % decay_step == 0 and i != 0:' by doing this, the learning rate never update.

the correct version I think should be 'if (i + epo * len(dataloader)) % decay_step == 0 and i != 0':

[Ugness]

I think you are right. I updated train.py. My code has a lot of out-dated coding style, and I recommend you use pytorch's lr-scheduler. I have a plan to update this repository with torch >= 1.4.0, but it would take some time. Thank you for making a comment.