Closed edmondja closed 7 years ago
Mostly because tffm is a research project focused on FMs It contains a lot of options which good for experiments and, imo, not suitable for Keras.
I believe it's not a good idea to make all tffm a part of Keras, but it may worth to implement PolyNet or 2-nd order FM as a layer (if not implemented yet)
Why don't you make your work compatible with Keras as a simple neural net layer ? https://github.com/fchollet/keras/issues/4959