Theano discontinuation
twiecki opened this issue · comments
What are your plans now that theano is going to be discontinued?
We, the pymc3 devs, have discussed all kinds of options. One of them is taking over theano maintenance (not pushing new features just making sure it doesn't go stale and critical bugs get fixed). Is that anything to talk about?
What are your plans now that theano is going to be discontinued?
As outlined in https://groups.google.com/forum/#!topic/lasagne-users/bldIYZ1UgKk, the plan would be to release the long overdue Lasagne 0.2, then aim for a final Lasagne 0.3 release that wraps up the open ends (some Theano 1.0 features are not covered yet, some parts might be refactored, some documentation could be added).
One of them is taking over theano maintenance (not pushing new features just making sure it doesn't go stale and critical bugs get fixed).
Keeping it working would be nice. I can't promise too much (some of the Theano internals are beyond my knowledge, and I won't have time to answer anything on theano-users or theano-dev), but I'd be happy to help every now and then.
I will add that currently Debian does include Theano and does not include Tensorflow due to some difficult packaging problems, so from this practical packaging standpoint I warmly encourage the continued support of a Theano backend for the moment, at least for 0.2.
@radarsat1 Debian ships theano https://tracker.debian.org/pkg/theano
Yes.
@f0k can you make a release for 0.2? I'd like to bump the version on conda-forge, given that lasagne 0.1 uses theano.tensor.signal.downsample, which was removed from theano after v0.7, but conda-forge only has theano>=0.8.2, so this obviously creates some problems when trying to use both in a conda env.