Vyas Ramasubramani
Vyas Ramasubramani
@gpucibot merge
@shaswat-indian @shwina I haven't looked at this code at all, but just FYI you may also benefit from looking at #9153
Based on discussion in #7492, we initially decided that the appropriate solution would be to match pandas by setting division by zero entries to zero in the output, but in...
I agree, I'd still want to do this as a post-processing step. I think we'd probably be OK to eat the perf hit though.
Yeah I think so. This feels like a very typical case where cuDF Python should pay a performance cost for matching pandas behavior exactly. No need to overspecialize libcudf.
This issue is pretty similar to #5938 in spirit, although it'll require a pretty different type of fix.
@gpucibot merge
rerun tests
CC @jlowe @revans2 for a Java perspective. AFAICT everything in the JNI just uses a single CudfException class. I don't know if there's any automatic way to generate a wider...
> > Probably the place to start is just by tacking a single pairing > > I think the place to start is to find where C++ exception `what()` messages...