Hello all,
Just wanted to let you know that I have released
version 0.4.0 of DeepLIFT which adds 1d convolutions and can convert sequential models trained with either Keras 1.1.2 or 0.3. It can also handle theano models that were trained with a tensorflow dimension ordering. Additionally, I have begun work on a tensorflow implementation of DeepLIFT which you can access on the
tensorflow branch. The tensorflow implementation passes all unit tests, but it has not been stress-tested.
I am debating whether to continue supporting both the theano and tensorflow backends, and am leaning towards pure tensorflow since it appears to be more popular in the deep learning community. If any of you have thoughts on this, I would love to hear them. Please also let me know if you have any other questions or feature requests.
Thanks,
Avanti