Tongfei Chen
Tongfei Chen
Thanks for your interest! This is partly working now -- garbage collection is not fully resolved. The java part has two part: one is purely generated from SWIG; the other...
Thanks @koen-dejonghe for the idea
Notes here from @kitsing : NumPy-style advanced indexing has form `A[X1, X2, ..., Xk]`, where - the indexee `A` has `k` dims, here in Nexus typed with axes-descriptor `(A1, A2,...
`G[_] = Batched[_]` naturally leads to Jacobian (#33).
Related to #16 .
No. Right now `nexus-diff` uses Wengert tape-based mechanism (the same as PyTorch). The paper above is a candidate for a future implementation. http://www.cs.cmu.edu/~wcohen/10-605/notes/autodiff.pdf -- this note on reverse-mode autodiff is...
Yes this is a Mac issue -- the standard installation of PyTorch does not support CUDA on Mac. You can either install PyTorch from source on Mac (
Seems that you need `-std=c++11` enabled in your C++ compiler on Mac? Probably putting that in `$(CXXFLAGS_Darwin)` in the `Makefile`?
@danyaljj Remove all CUDA references in your `Makefile`. PyTorch on Mac, unless built from source, does not have CUDA support.
That's right -- I don't want to use `autoconf` (`.configure` then `make`) -- that's too much hassle. If you'd like you can modify the `Makefile` to check if `$(CUDA_HOME)` is...