separate requirements for logger dependencies

[ghost]

Fixes #764.
The versions are the current versions of the respective packages.
I used all loggers in offline mode and they all seemed to work. Not sure how exhaustive the test should be.

[williamFalcon]

thanks! why do we need a separate requirements? we already have a different one in tests.

[ghost]

ah thanks didn't know about that. The problem is, that we need to specify versions of optional packages (e.g. loggers) somewhere. More as a reference for users than a file for installing requirements.

[Borda]

@williamFalcon as a user you do not want to install all testing packages, just those you need to run e.g. special logger... we have a warning message if you ale loading logger for which you do not have a package installed, but it is missing min version...

[Borda]

pls copy versions from tests/requirements.txt with minimal versions
then in tests/requirements.txt imports these extras like -r ../requirements-extra.txt

[ghost]

@Borda only test_tube had a minimal version. Copied that, added -r ../requirements-extra.txt and removed the loggers from test/requirements.txt

[Borda]

@Aljo-Rovco it would be great contributions to find out what is the min versions to pass all tests (you can try something like divide an conquer strategy =)
You even do not meet to run all tests, just tests/test_loaders.py

[ghost]

will deal with this either today or tomorrow

[ghost]

@Borda I assume you meant tests/test_logging.py?

[ghost]

I increased the amount of data that is used when running the tests (from 0.01 to 0.05), because otherwise the log method in the wandb-logger is not called. And if that method is not called, the tests would run fine with wandb==0.8.14. I increased the number for the other loggers as well, just to be on the save side

[ghost]

@Borda could you have a look at the failing checks? Not quite sure how to fix them

[Borda]

lists of files in version control and sdist do not match!
suggested MANIFEST.in rules:
  include *.txt
missing from sdist:
  requirements-extra.txt

@Aljo-Rovco you need to include or exclude the new file in MANIFEST.ini

[Borda]

it seems there is an ussie with wandb may you have look at it...

wandb.run_manager.LaunchError: W&B process failed to launch, see: C:\Users\RUNNER~1\AppData\Local\Temp\wandb-debug.log

[ghost]

had a look, but not sure how to solve it.
line 1126 in wandb/run_manager.py (stdout_read_file = os.fdopen(stdout_read_fd, 'rb')) throws OSError: [WinError 6] The handle is invalid. Not sure why. Could it be the location of the tempdir that is used for logging?
I don't have access to a windows machine, so I cannot recreate/test it.

[Borda]

it was working before so let's try to increase the wandb version...

[ghost]

I think it was working before because the test didn't test much. It just created a logger:

def test_wandb_logger(tmpdir):
    """Verify that basic functionality of wandb logger works."""
    tutils.reset_seed()

    wandb_dir = os.path.join(tmpdir, "wandb")
     _ = WandbLogger(save_dir=wandb_dir, anonymous=True) 

All other tests actually trained a model

[ghost]

also, isn't the last version used in the CI testing?

[Borda]

also, isn't the last version used in the CI testing?

add line in .github/workflow/ci-testing.yml

python -c "req = open('requirements-extra.txt').read().replace('>', '=') ; open('requirements-extra.txt', 'w').write(req)"

just belo the changing requirements.txt so we have the min version propoerly tested

[Borda]

I think it was working before because the test didn't test much. It just created a logger:
...

I would say you can try to fix it here or remove the new test and open issue where you describe the problem along with your proposed test... I would suggest to to to the second option so we keep the PR simple... sure you can take the new issue and solve it :] @Aljo-Rovco

[Borda]

LGTM 🚀

[Borda]

@Aljo-Rovco pls create an issue with suggesting the tests you had here for wandb so we do not forget it and keep everything well tested :]

[ghost]

Sorry, was busy. Will do

[ghost]

created a new issue: #906