problem with the normalize weights function for convolution layer in tensorflow branch about deeplift HOT 4 CLOSED

Yes that is very likely the problem. I can update that, thanks for pointing it out.

from deeplift.

Hi @satroz, do you mind letting me know how soon you need this? I am planning a significant update to deeplift in the next week or so (which will be accompanied by a new arXiv preprint), and I would ideally like to bundle this in with the update. However, if you need it soon, let me know and I will prioritize that.

from deeplift.

Thanks ! Yeah no worries, I guess I can wait another week or two for the big update.... but if you think it'll take longer than that then would be nice to have it sooner then ;)

from deeplift.

Hey @satroz, big update coming soon (need to get approval from my adviser to release it) but in the mean time I just pushed an update to the current tensorflow branch that lets you specify the axis of normalisation. I tested it out on a model with tensorflow ordering (weights of first conv layer had shape 4xx1x) and it seems to work - you can view the notebook with the test here: https://github.com/AvantiShri/model_storage/blob/4143cfce7e61611d4c42984578e420cd7556c4b6/deeplift/genomics/test_mean_normalise_weights.ipynb

Let me know if you run into any trouble. I am hoping to push the big update out by Tuesday, pending a green light.

from deeplift.