targets are not really copied when using cpu backend
loofahcus opened this issue · 1 comments
loofahcus commented
deepqnetwork.py#L133 should be:
targets = preq.asnumpyarray().copy()
Since for cpu backend, asnumpyarray()
seems just get the reference of numpy array, which causes the modifications of targets
impacting preq
.
tambetm commented
Thanks @loofahcus! I was aware of this, but forgot to include the fix in this repo.