search

aigamedev / scikit-neuralnetwork

Deep neural networks without the learning cliff! Classifiers and regressors compatible with scikit-learn.
BSD 3-Clause "New" or "Revised" License
1.21k stars 221 forks source link

Future-proof NN interface with multiple backends #57

Closed alexjc closed 9 years ago

alexjc commented 9 years ago

Noticing how fast the deep learning ecosystem is moving, with multiple NN implementations in Python, and much of the code in Lua, @ssamot and I are considering moving the project toward a multiple back-end interface with a stable interface — in order to keep things as future-proof as possible. This seems like one of the best ways to support Deep Learning in Python too.

In particular, we're considering keras.io as a new backend as it supports some interesting RNN features.

What do you guys think?

alexjc commented 9 years ago

From version 0.3 onward, we'll be taking the library in this direction. @ssamot has a branch called RNN with a passage integration, for example. The focus will remain on a simple to use library with a stable API for end-users rather than neural network experts.

alexjc commented 9 years ago

There's a new branch called backend which has my refactoring for supporting multiple neural network libraries with the same compatible interface. Started with deepy for no particular reason, keras is another great option we'll follow up with once this stabilises.