Jonathan Malmaud
Jonathan Malmaud
A Cxx docker image exists: malmaud/julia:cxx
Thanks for the report!
Ya, this would be great. It's relatively low-priority for me at the moment, so not sure when I'll be able to address it myself.
It's being slowly worked on, but isn't available yet. You can import a saved Python metagraph, though.
Not at the moment, although it should be pretty straightforward to add.
No. There is a just TODO in the readme for the C API that says to add support support for gradient calculations.
I'd be up for lazy-loading the python process. In practice I would think most uses of TensorFlow would use `gradients`, but if nothing else, you'd get a faster start-up time.
It is is closer, but still far away - many gradients are still missing from the C API.
How about a compromise for now where all functions related to devices take a keyword argument for whether to interpret them as one-base or zero-based? I don't want to break...
Ya you're right. Want to submit a PR?