Tensorflow independent environment

[VIGNESHinZONE]

Fix #2561

• I have added try & catch based on a few sub-packages in deepchem like metalearning, keras_models, rl because these sub-packages are tensorflow dependent.
• Didn't have a lot of problem in trans subpackage because the TensorFlow dependency was local.

cc: @peastman @rbharath

#2563 requires a tensorflow free environment for smoothly creating a deepchem-[torch] environment.

[VIGNESHinZONE]

In transformer subpackage, only IRVTransformer & ANITransformer have tensorflow depedency.

deepchem/deepchem/trans/transformers.py

Line 1540 in eab6ee6

class IRVTransformer(Transformer):

deepchem/deepchem/trans/transformers.py

Line 1977 in eab6ee6

class ANITransformer(Transformer):

Can we move these features into a seperate file called tensorflow_transformers and eventually do this in __init__.py

try:
  from deepchem.trans.transformers import IRVTransformer
  from deepchem.trans.transformers import ANITransformer
except:
  pass

[peastman]

It looks to me like they're just using TensorFlow for a bunch of standard math operations that could just as easily be done with Numpy. I suggest we convert them.

I notice though that there don't seem to be any tests for ANITransformer, and also that it's written in TensorFlow 1.x style. I wouldn't be surprised if it doesn't actually work, but no one noticed due to the lack of tests and because no one is using it.

[codecov-commenter]

Codecov Report

Merging #2567 (21dde75) into master (ab4bdff) will increase coverage by 0.31%.
The diff coverage is 78.37%.

@@            Coverage Diff             @@
##           master    #2567      +/-   ##
==========================================
+ Coverage   84.85%   85.17%   +0.31%     
==========================================
  Files         314      314              
  Lines       28084    27972     -112     
==========================================
- Hits        23831    23824       -7     
+ Misses       4253     4148     -105     

Impacted Files	Coverage Δ
deepchem/trans/__init__.py	100.00% <ø> (ø)
deepchem/metalearning/__init__.py	50.00% <50.00%> (-50.00%)	⬇️
deepchem/rl/__init__.py	63.93% <60.00%> (-1.59%)	⬇️
deepchem/models/__init__.py	85.36% <84.00%> (-8.92%)	⬇️
deepchem/trans/transformers.py	86.09% <100.00%> (+13.31%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ab4bdff...21dde75. Read the comment docs.

[VIGNESHinZONE]

So should i put ANITransformer in a separate file like mentioned above or remove it completely?
removed tf dependency from IRVtransformer

[rbharath]

@VIGNESHinZONE I think for ANITransformer, it's been known to be broken for some time so let's just comment it out. We'll have to note it in #2553.

[rbharath]

I think this PR will also need to be rebased since some merge conflicts have crept it

[VIGNESHinZONE]

@rbharath I have made the changes.

[rbharath]

Ok, I think this is ready to merge in. I'm seeing only 1 test failure on a test unrelated to this PR.

@peastman Do you have any remaining review comments?

[peastman]

Looks good to me.