Documenting the New CI fixes

I have made changes in Github Actions, which provides four separate packages in deepchem for the different environments -

1. Deepchem
2. Deepchem[tensorflow]
3. Deepchem[torch]
4. DeepChem[jax]

Changes have been made in this https://github.com/deepchem/deepchem/pull/2573 . Its a long PR, and it has become difficult to manage it with new errors showing upon new PR’s being merged into Master branch and hence I was asked by @rbharath @ncfrey to document all the errors showing so that we could manage them more easily.

Main Feature of PR

1. Separate test suites for different environments using Pytest Markers - tensorflow, torch, jax. Along with that, had to make import changes accordingly
2. Made changes in Github Actions accordingly and have added scripts for installing these packages separately.
3. There four environments in CI - common(all those which are only numpy or scikit-learn dependent) , tensorflow, torch, jax.
4. Had to fix bugs accordingly, based on each individual environment.

Unfixed Errors from the PR -

1. DeepChem Common -

Most of these errors are interconnected, fixing one might solve the whole bucket

• Errors in FastaLoader -

• TestFASTALoader.test_fasta_one_hot
• TestFASTALoader.test_fasta_one_hot_big
• TestFASTALoader.test_legacy_fasta_one_hot

• Errors of the form

TypeError: join() argument must be str or bytes, not 'float64'

• deepchem/data/tests/test_data_loader - error log
• deepchem/data/tests/test_datasets.py - error log
• deepchem/trans/tests/test_flattening.py error log
• deepchem/trans/tests/test_normalization.py error log

• Errors in Gaussian HyperParameter - I was not able to reproduce the errors in my local environment or google collab notebook

• deepchem/hyper/tests/test_gaussian_hyperparam_opt - error log

2. DeepChem Tensorflow -

Our test suite has also included slow tests (which were previously checked only before release cycles),

• Flaky errors, which are inconsistent in Python 3.6, 3.7

• AtomicConvFeaturiser in deepchem/feat/tests/test_graph_features.py - error log
• WeaveModel in deepchem/models/tests/test_weave_models.py - error log

2. DeepChem Torch -

Our test suite has also included slow tests (which were previously checked only before release cycles),

3. DeepChem Torch -

Same as tensorflow, we are running slow tests in torch environment as well and here is the one main flaky failing test-

• MultiTaskClassification on qm7 dataset. error log

4. Few More errors in DeepChem Common

Error occurring only in Python 3.8 and 3.9 . These errors seem related to OpenVino and the Docking engine

• TestBindingPocket.test_convex_find_pockets error log
• TestBindingPocket.test_extract_active_site error log
• TestBindingPocket.test_get_face_boxes_for_protein error log

And quite a few more which can be found in the same logs related to the Docking.

[peastman]

A lot of these errors were fixed by #2605. I checked them off above.

[peastman]

@ncfrey the error in AtomicConvFeaturizer was introduced by #2406. It added a new required positional argument n_tasks to AtomicConvModel. This broke AtomicConvFeaturizer, which constructs an AtomicConvModel. Could you take a look and see what you think the appropriate fix is? At the point where it creates the model, I don’t think we know yet how many tasks there are going to be. Perhaps we need to wait until later to create the model?