soft-decision-tree icon indicating copy to clipboard operation
soft-decision-tree copied to clipboard

Published 20 hours ago • verified publisher icon kimhc6028

Error when learning rate is big

Open f-cg opened this issue 5 years ago • 2 comments

python3 main.py --max-depth 4 --lr 1 ... Train Epoch: 3 [17280/60000 (29%)] Loss: 0.522095, Accuracy: 54/64 (84.0000%) Traceback (most recent call last): File "main.py", line 79, in model.train_(train_loader, epoch) File "/content/model.py", line 190, in train_ correct += pred.eq(target.data).cpu().sum() RuntimeError: Expected object of backend CPU but got backend CUDA for argument #2 'other'

This is so stange and I can't fix it.

f-cg avatar May 07 '19 16:05 f-cg

do you need high lr? does this bug occur when lr is relatively small? lr=1 seems like unlikely

kimhc6028 avatar May 08 '19 01:05 kimhc6028

Ok, I don't need that high lr, but this issue can happen when lr is smaller and epoch is bigger.

f-cg avatar May 08 '19 03:05 f-cg