I wanted to use torch.optim.lr_scheduler.OneCycleLR
while training. Can some kindly explain to me how to use.
What i got from the documentation was that it should be called after each train_batch
.
My confusions are as follows:
-
Does the
max_lr
parameter has to be same with theoptimizer
lr
parameter? -
Can this scheduler be used with
Adam
optimizer. How is the momentum calculated then ? -
Let’s say i trained my model for some number of epochs at a stretch now, i wanted to train for some more
epochs
. Would i have to reset the thescheduler
?
Can anybody provide me a sort of a toy example/traininng loop that implements this scheduler ? Any help would be appreciated ?
P.S: I am kind of new to deep learning
& PyTorch
so my question might be somewhat silly.