Requirements fight over numpy version

[jasper-tms]

Currently pip install numpy, pip install -e . fails with some enormously long error message about struggling to resolve dependencies and failing to build numpy. Looks like pip somehow comes to the conclusion that it wants to build numpy version 1.13.1 from source, but it can't. The root of the problem may be deepinterpolation, because if I remove that from requirements.txt, the install succeeds with no issues. If I first pip install https://github.com/NeLy-EPFL/deepinterpolation/archive/adapttoR57C10.tar.gz and then pip install -e . with deepinterpolation removed from twoppp's requirements, the install succeeds, although with the two warnings:

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
tensorflow 2.3.0 requires numpy<1.19.0,>=1.16.0, but you have numpy 1.23.3 which is incompatible.
deepinterpolation 0.1.1 requires numpy==1.16.0, but you have numpy 1.23.3 which is incompatible.

so it looks like the install ends up keeping recent numpy, which I think is a good thing. I have not yet tested whether the twoppp install performed this way (first deepinterpolation, then the rest) actually works, but if it doesn't, the solution shouldn't be to downgrade numpy but instead to upgrade deepinterpolation and/or tensorflow to work with new numpy.

I will note that the latest version of the Allen Institue's deepinterpolation requirements.txt file specifies tensorflow==2.4.4 and numpy with no specific version number, so we could try to get our deepinterpolation fork updated to have those changes made in the Allen's repo.

[jasper-tms]

@VictorLobatoRios and I have both run deepinterpolation with tensorflow 2.4 recently, so indeed it sounds like we can change https://github.com/NeLy-EPFL/deepinterpolation/blob/master/requirements.txt and resolve the dependencies issue.

Coincidentally, the Allen Institute's version of deepinterpolation was updated just yesterday to list tensorflow==2.7 instead of tensorflow==2.4.4, so we could probably even start using that.