dim parameter for softmax
Is it possible to specify the dim parameter for softmax and similar functions, as available in the python api? https://www.tensorflow.org/versions/r0.11/api_docs/python/nn.html#softmax
Not at the moment, although it should be pretty straightforward to add.
Sorry for the lack of effort from my part. I would like to help, but when I tried to read the code I didn't quite understand how to add this functionality. When looking at the tensorflow python api it seems as they do a transpose, do the computation, and then transpose back if you want to do computations in a different dimension. I find that you can gain a lot of performance when using sample index as the last dimension instead of the first. Especially when using mini batches.