[JOSS] Tests

[RMeli]

The only test in the test folder fails with the following error:

Warning: importing 'simtk.openmm' is deprecated.  Import 'openmm' instead.
Dataset.shape: torch.Size([16, 3, 2145])
mean: 44.69796567599069, std: 13.886234009167817
E
======================================================================
ERROR: test_openmm_energy (test_openmm_plugin.Test_openmm_plugin.test_openmm_energy)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/rmeli/git/molearn/test/test_openmm_plugin.py", line 30, in test_openmm_energy
    openmmscore = molearn.openmm_energy(data.mol, data.std, clamp=None, platform = 'CUDA') #xml_file = ['modified_amber_protein.xml',])
                  ^^^^^^^^^^^^^^^^^^^^^
AttributeError: module 'molearn' has no attribute 'openmm_energy'

----------------------------------------------------------------------
Ran 1 test in 0.062s

FAILED (errors=1)

While the functionality of the software can verified through the provided example notebooks and the scripts in the examples folder, I think the code could greatly benefit from some additional unit tests (and, eventually, the use of CI).

This relates to my comment in #8; if tests, examples scripts, and notebooks are not run each time the main package is updated they are bound to diverge eventually.

---

Since the example_script.py is not testing anything (as far as I can tell), should it be moved elsewhere (for example the examples folder)?

---

The only necessary action item here (IMO) is to fix the only test available.

From my understanding of JOSS criteria, building a more extensive test suite and adding CI is not required. Personally I'd strongly recommend it as it will benefit future development, but can be added in the future.

---

https://github.com/openjournals/joss-reviews/issues/5523

[degiacom]

Thank you for your feedback, I agree testing needs to be improved. We have now updated the unittests in test/test_openmm_plugin.py. Concerning example_script.py, it is currently used in the CI of conda-forge.

[RMeli]

Thanks for updating the test and adding some more! I was able to run them with python -m coverage run -m unittest and they all pass.

Ran 32 tests in 7.462s

OK

Coverage report:

Name                                                                               Stmts   Miss  Cover
------------------------------------------------------------------------------------------------------
/home/rmeli/git/molearn/src/molearn/__init__.py                                        5      0   100%
/home/rmeli/git/molearn/src/molearn/data/__init__.py                                   2      0   100%
/home/rmeli/git/molearn/src/molearn/data/pdb_data.py                                 122     25    80%
/home/rmeli/git/molearn/src/molearn/loss_functions/__init__.py                         3      0   100%
/home/rmeli/git/molearn/src/molearn/loss_functions/openmm_thread.py                  262    121    54%
/home/rmeli/git/molearn/src/molearn/loss_functions/torch_protein_energy.py           213    191    10%
/home/rmeli/git/molearn/src/molearn/loss_functions/torch_protein_energy_utils.py     761    740     3%
/home/rmeli/git/molearn/src/molearn/trainers/__init__.py                               5      0   100%
/home/rmeli/git/molearn/src/molearn/trainers/openmm_physics_trainer.py                39     28    28%
/home/rmeli/git/molearn/src/molearn/trainers/sinkhorn_trainer.py                     206    170    17%
/home/rmeli/git/molearn/src/molearn/trainers/torch_physics_trainer.py                 38     27    29%
/home/rmeli/git/molearn/src/molearn/trainers/trainer.py                              202    169    16%
test_openmm_plugin.py                                                                138      1    99%
------------------------------------------------------------------------------------------------------
TOTAL                                                                               1996   1472    26%