
I need something that I can reproduce paper results and experiment myself with different architectures and tuning, extracting embeddings and that sort of thing. I know of a couple of frameworks such as TF, TF-Learn, TF-Slim, Theano so I'd like suggestions on which ones I should use.