0.17.5 RCs do not detect scikit-learn dependency

[odedbd]

I've been testing 0.17.5rc0-0.17.5rc2 and scikit-learn dependency isn't being detected. As a result my scripts, which used to run fine on remote, fail on import sklearn. I verified that indeed scikit-learn is not listed in the dependencies for the task. Reverting to 0.17.4 resolved this issues.

[bmartinn]

Thanks @odedbd
I was able to locate the issue, a fix will be ready soon, I'll update here.
BTW: This will only affect scikit-learn, since the import is different from the package naming :)

[odedbd]

@bmartinn that's great. Is there an option to manually register dependencies, in a addition to the automagical discovery? Some API like task.add_dependency('package_name',version=None) where version is optional if you want to just grab the latest? With such an API I could fix such a problem on my end, at least until it would be solved "properly".

[jkhenning]

Hi @odedbd ,

Is there an option to manually register dependencies, in a addition to the automagical discovery? Some API like task.add_dependency('package_name',version=None) where version is optional if you want to just grab the latest?

Yes indeed, you can use:
task.add_requirements(package_name, package_version=None)

🙂

[odedbd]

That's good to know 👍

[bmartinn]

@odedbd
Updating here the latest RC fixes this issue, please feel free to test:

pip install clearml==0.17.5rc3

[odedbd]

Working, thanks!

[graelo]

Hi,

For future readers, the api for manually registering dependencies is currently a class method (note the Task, not task)

Task.add_requirements("requests", "2.29.0")

and this statement must be placed before Task.init(...)