Closed foster999 closed 4 months ago
Thanks for reporting this error and providing a reproducer. I could identify the issue and created a PR to fix it. In the meantime, you could use a normal NeuralNetClassifier
with two classes, which works the same as NeuralNetBinaryClassifier
for most practical purposes, and should have no issues with torch.compile
.
I'm using Python 3.11.3, skorch==1.0.0
I find the error disappears when dropping the
compile
argument. It doesn't seem to error for similar examples withNeuralNetClassifier
.Minimal example
Raises