Pycaret Library Note working. Suspected multithread issue.

[ajc]

Hi Knime Team,

I have been playing around with pycaret (PyCaret — pycaret 3.0.2 documentation) and Knime. I have come across an issue when using the create_model and compare_model api’s. The issue appears to be knime related as they are working in a jupyter notebook.

Further more the issue appears to be related to the n_jobs parameter set in the setup() api.

n_jobs: int, default = -1
The number of jobs to run in parallel (for functions that supports parallel processing) -1 means using all processors. To run all functions on single processor set n_jobs to None.

https://pycaret.readthedocs.io/en/latest/api/regression.html

When set to -1 or greater than 1 the api’s fail. When set to 1 the apis work however this is not ideal as there is a significant performance impact.

I have attached example errors raised and a workflow to demonstrate the problem further.

Any help with this one would be greatly appreciated. Knime is an awesome tool and will be even better with pycaret!

Cheers
pycaret_error.knwf (14.1 KB)

[Andrew_Steel]

Hi @ajc,

it works fine for me …

This points to differences in the Python installation. I have installed conda 4.9.2 and Python 3.6.13 and pycaret 2.3.0.

Best regards
Andrew

[ajc]

Thanks for checking on this Andrew.

It seems it is a python version issue. I was using python 3.8.5. A quick test using python 3.6.13 proved successful.

Thanks again

[ben_westphal]

Andrew, just stumbled across your post. We should catch up, Im keen to know how you have progressed. My number is 0424 993 413. Cheers, Ben. Forest Grove.