Bump scikit-learn version to 1.6

[ArturoAmorQ]

We currently use scikit-learn v1.3, but #772 requires v1.4. Updating scikit-learn to its latest version (v1.6) will allow us to give a better maintenance to the mooc.

On FUN, updating the Jupyter image of the students seems to suffice.

[lesteve]

Not sure about the implications on the FUN MOOC if any, but I'll let you deal with this 😉

[lesteve]

Thinking about it more, I guess for now none of our notebooks actually need scikit-learn 1.6 but oh well ...

[ArturoAmorQ]

Not sure about the implications on the FUN MOOC if any, but I'll let you deal with this 😉

Pinging @brospars for the Jupyter image update :)

Thinking about it more, I guess for now none of our notebooks actually need scikit-learn 1.6 but oh well

Not necessarily 1.6, but #772 needs 1.4, (until now, we used to run on 1.3)

[lesteve]

Pinging @brospars for the Jupyter image update :)

I seem to remember in FUN-MOOC the environment is created when the user starts its first notebook. I don't remember if there was a way for the user to update its environment but @brospars is indeed the one that may know 😉.

until now, we used to run on 1.3

This does not matter too much but just to be 100% explicit. 1.3 is our minimum version requirement scikit-learn>=1.3 meant that we were using the latest release, for example in the CI or if someone new come to the repo and follows our installation instructions.

[brospars]

The notebook image is this one : https://hub.docker.com/r/brospars/singleuser-scikit-learn-mooc (source : https://github.com/inrialearninglab/singleuser-scikit-learn-mooc)

I disabled auto updates last year to prevent potential errors due to breaking changes.

If I run the pipeline to update dependencies. This will run these lines :

FROM jupyter/scipy-notebook:notebook-6.5.4
# [...]
# Install mooc requirements
RUN wget https://raw.githubusercontent.com/INRIA/scikit-learn-mooc/master/requirements.txt
RUN python3 -m pip install --no-cache -r requirements.txt

And every user will have the updated version upon the next startup of their environment.

Do I proceed ?

[lesteve]

I would wait at least for scikit-learn 1.6.1 to be released, which will happen soonish (I have not followed closely but I would say a week or so).

About the answer to "should we update the scikit-learn version in the MOOC JupyterHub?", the conservative part of me is a bit 🤔 "if it's not broken, don't fix it".

The adventurous part of me is like "once scikit-learn 1.6.1 is released why not, what could possibly go wrong?". Having said that, once the change is done in the MOOC JupyterHub, it would be nice to test a few notebooks manually and make sure everything seems to work smoothly.

[brospars]

About the answer to "should we update the scikit-learn version in the MOOC JupyterHub?", the conservative part of me is a bit 🤔 "if it's not broken, don't fix it".

That was exactly my thought 😅 but 1.4 seems to be required by #772

it would be nice to test a few notebooks manually and make sure everything seems to work smoothly.

Yes I will do that first on my eprsonnal env and then deploy it for everyone

[brospars]

Update : all "solution notebooks" seems to be running fine on 1.6.0. Once 1.6.1 is out I will deploy it.

[lesteve]

Update : all "solution notebooks" seems to be running fine on 1.6.0. Once 1.6.1 is out I will deploy it.

Thanks for checking 🙏! By "solution notebooks" you mean with those with _sol_ in their names or wrap-up quiz solutions? I think wrap-up quiz solutions would be good candidate notebooks to test indeed.

Maybe @ArturoAmorQ has some suggestions for some computationally heavy notebooks that would be good to test as well.

[brospars]

I meant _sol_ notebooks

[ArturoAmorQ]

Thanks for the check! I'll have a run of the wrap-up quizzes to confirm everything works as expected with 1.6. I'll ping you when it's done @brospars.