Tensorflow to pytorch is now in the dev branch

[pgunn]

Hello,
Thanks to a lot of hard work from @manuelpaeza we now have, in the dev branch, a version of Caiman that no longer uses Tensorflow, instead depending on pytorch. This will allow for the next release to be properly packaged on Windows again, and will make things a little easier packaging/install-wise for all Caiman users.

I'm also going to be working on fixing several more packaging issues and notebook problems in the codebase before the next release; a few of the notebooks haven't worked correctly for awhile.

If any community members want to help with testing what's in dev, please do so and let me know how it goes - try the cli and notebook demos, try your scripts. Note that you'll need to do a new install of caiman - the dependencies have changed. I'm hoping to be able to do a new release within the next two weeks, assuming I can get all the biggest issues solved within that timeframe.

[katonage]

Hi Pat! When it is expected to release this update to the conda-forge? (Getting caiman from conda-forge works on one windows machine while the estimates.evaluate_components fails on the other windows machine. something related to tensorflow threading...)

[pgunn]

@katonage
Right now we're working through some problems with components evaluation being very broken with the pytorch version; there are some other things I need to fix before we can release and it'll need a lot of testing in general. Might be a bit - sorry!

We're very eager to get this out the door, when it's ready. The Windows release of Caiman is a few versions old.

[pgunn]

Hello,
Manuel and I have both spent a lot of time fixing some of the bumpiness with this (Manuel deserves most of the credit), and things are in a much better shape; I'm presently going through the notebooks and trying to correct outstanding issues (many unrelated to the patch) to get the code into a good state before our next release; if you tried it before and hit bumps, please try it again.