Documenting the New CI fixes
See original GitHub issueI have made changes in Github Actions, which provides four separate packages in deepchem
for the different environments -
Deepchem
Deepchem[tensorflow]
Deepchem[torch]
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
- Separate test suites for different environments using Pytest Markers -
tensorflow
,torch
,jax
. Along with that, had to makeimport
changes accordingly - Made changes in Github Actions accordingly and have added scripts for installing these packages separately.
- There four environments in CI -
common
(all those which are onlynumpy
orscikit-learn
dependent) ,tensorflow
,torch
,jax
. - 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
indeepchem/feat/tests/test_graph_features.py
- error log -
WeaveModel
indeepchem/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
onqm7
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.
Issue Analytics
- State:
- Created 2 years ago
- Reactions:2
- Comments:11 (11 by maintainers)
Top GitHub Comments
A lot of these errors were fixed by #2605. I checked them off above.
@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?