New loguru logging seems not to play nice with Python logging.

[sarusso]

Hi,

commit 61d9845 introduced loguru, but it seems to me that right now it's just using Python root logger under the hood, as it might not be properly set up (as mentioned in https://github.com/Delgan/loguru?tab=readme-ov-file#suitable-for-scripts-and-libraries)

The result is that it becomes hard to include Fitter v1.7.0 in other applications that might need to change logleves, use other handlers or just hide its output. In this sense, it was better to use Python's standard logging.

I am not sure what the solution could be, but right now I am forced to monkey-patch Fitter at runtime to be able to take control over its logging.

[cokelaer]

dear @sarusso thanks for reporting this issue. indeed, I naively used loguru and I believe this is not really what I wanted. I need to look a bit more at the doc you sent and will probably roll back to previous version.

[yinchi]

I am using Fitter in a Python notebook and cannot seem to disable the warning messages no matter what I do. If anyone has the solution, it would be much appreciated.

[chrisschopp]

@sarusso Would you be willing to share how you monkey-patched Fitter to control the loguru logging?

[sarusso]

@sarusso Would you be willing to share how you monkey-patched Fitter to control the loguru logging?

Sure, I just did the following:

import logging
import fitter

fitter.fitter.logger = logging.getLogger('fitter')
fitter.fitter.logger.setLevel(level=logging.CRITICAL)

Note however that this does not control the loguru logger, it just replaces it altogether with the standard Python logger.

[yinchi]

I created my own fork to solve the problem, which seems to even play well with multiprocessing and joblib (that was a challenge!). This requires setting the logging level for each subprocess using a parameter passed into the Parallel call. I don't think @sarusso's solution above addresses this part.

Not submitting a PR for now, as I changed other stuff as well (including requiring Python 3.12, which is probably not a good thing if others need to use the library). If @cokelaer wants to merge this, I can try creating a version that only changes the fitter.py code and keeps the poetry setup the same as before.

https://github.com/yinchi/fitter

[srimam]

Quick note: to successfully use the solution provided by @sarusso I had to limit the number of max_workers to 1 (consistent with @yinchi comment.

f.fit(max_workers=1)

This along with the lines from code from @sarusso suppressed the info level logging for me.