Python 3.7 Support

[laserprec]

Description

We should extend the support of recommenders to python 3.7.

Expected behavior with the suggested feature

We should modify our CIs to run our checks in py37 and fix any relevant issue raised from it.

Other Comments

@anargyri, I think you mentioned problems with numba in python3.7 in venv installation. Is there an existing issue we can link to? I can also take a look.

[anargyri]

I don't think we have opened an issue about this. But I found a workaround, see https://github.com/microsoft/recommenders/tree/andreas/simon_feedback#getting-started
i.e. if you conda install numpy-base at the end then the tests pass.

[anargyri]

I could not get venv to work with python 3.7

[laserprec]

My understanding is that a venv in a conda-enabled environment (including a DSVM) is not truly self-contained. I have tried venv installation on a DSVM (without conda activated) and run into sklearn-surprise installation issues and this is not an issue if I run venv installation in a fresh ubuntu image... Maybe there's some custom PATH/ENV VAR modification in DSVM so it's not so friendly for venv? I can try to replicate this error in 3.7 on an ubuntu image.

[anargyri]

Well, I have tested venv on a DSVM and it does work with Python 3.6 (see these docker images https://github.com/microsoft/recommenders/blob/4a9063d6ba1f5216816661516f29cbff3494879c/tools/docker/Dockerfile)
The breaking issue between 3.6 and 3.7 is that numpy error (which conda addresses with numpy-base).

[laserprec]

Yup. I can replicate this import error in py3.7 in venv.

.tox/gpu/lib/python3.7/site-packages/numba/core/pythonapi.py:12: in <module>
    from numba import _helperlib
E   ImportError: numpy.core.multiarray failed to import

[anargyri]

@laserprec I found a solution to this (see the above PR) by installing Surprise from source (the issue was caused by the .pyx files in Surprise which apparently were compiled with an incompatible numpy version).

[laserprec]

@anargyri this is AWESOME! Thanks Andreas!