Training fails with IndexError
when performing train/test dataset splitting
#104
Labels
bug
Something isn't working
Running a simple training command with some low-resolution training data (only 100 samples) gives me an error that gets triggered in
train_for_one_epoch
, when weenumerate(dataloader)
. Onmain
branch, runOn main (as of the merging of #97 in early December 2023)
gives
IndexError: index <x> is out of bounds for axis 0 with size 80
. The index seems between 80-320 (I've certainly seen low 80s and high 300s).There are 320 samples in all training subdomains combined (4 spatial domains, 80% each for training). We do batching with a size of 4. I've tried investigating and tinkering with these, but I've not managed to resolve it.
Doing either of these prevents the issue from occurring:
It would seem the problem is somewhere in
Subset_
or related code, or in my xarray generated from the data step.The text was updated successfully, but these errors were encountered: