hi there, when i read the code of train.py, i notice this in line 127, 128:
if epoch % args.writing_interval == 0:
print(f'|-Train-|{epoch}: {train_loss:.3f}')
writer.add_losses("Loss", "train", train_loss, epoch)
writer.add_losses("Loss_l1", "train", train_loss, epoch)
so i am wondering is it that you want to record loss_l1 but used the wrong "train_loss"?
also you guys have mentioned in your paper that, during training you randomly cropped 5-second segments, but in your code the default win_length is 10 and thus the default length is 10*64=640, so i am kind of confused. i would really appreciate it if you can give me some insights about this win_length.
hi there, when i read the code of train.py, i notice this in line 127, 128: if epoch % args.writing_interval == 0: print(f'|-Train-|{epoch}: {train_loss:.3f}') writer.add_losses("Loss", "train", train_loss, epoch) writer.add_losses("Loss_l1", "train", train_loss, epoch) so i am wondering is it that you want to record loss_l1 but used the wrong "train_loss"?
also you guys have mentioned in your paper that, during training you randomly cropped 5-second segments, but in your code the default win_length is 10 and thus the default length is 10*64=640, so i am kind of confused. i would really appreciate it if you can give me some insights about this win_length.