Mailman 3 May 2026 - scikit-learn

Re: invitation to present at the SciPy 2026 Tools Plenary
by Adrin May 26, 2026

May 26, 2026

Hi Inessa, Thanks for the invitation. I won't be able to attend/present. But sharing this with the team (cc) to see who can attend. Thanks, Adrin On Mon, May 25, 2026, 20:20 Inessa Pawson <inessa(a)albuscode.org> wrote: > Hi, Andreas, Adrin, and Reshama! > This is an invitation for the scikit-learn team to share your project > updates at the SciPy 2026 Tools Plenary. > > Given that these presentations are limited to just 3 minutes (yes, even > shorter than a lightning talk!), we recommend preparing only 2–3 slides > that provide a brief project overview, highlight new features and/or > upcoming changes, showcase community activities, and share opportunities > for contribution. It's important to note that your talk must be delivered > in person. > > The Tools Plenary sessions will take place during the main conference, > Wednesday, July 15 - Friday, July 17. A detailed schedule, including your > exact time slot, will be shared closer to July 15. If you have a preference > for which day you would like to present, please let us know as soon as > possible so we can do our best to accommodate your schedule. > > *Please RSVP to this invitation by Monday, June 1.* > > Best regards, > Inessa Pawson and Sanket Verma > SciPy’26 Tools Plenary Chairs >

1 0

ANN: Compute! Paris 2026 conference - Call for Proposal extended
by Guillaume Lemaître May 25, 2026

May 25, 2026

Hi all, The team of JupyterCon 2023, PyData Paris 2024 & 2025 organizes a new conference named Compute! Paris 2026 on open source computation and data. The conference will convene at Sorbonne Université in Paris on November 25–26, 2026. While the event maintains a broader focus than its predecessors and encompasses various computational domains, Python-centric content is anticipated to feature prominently given the language's widespread adoption. In 2026, neither JupyterCon nor PyData conferences will be held in Paris, so join us at Compute! Paris. CfP deadline has been extended to June 7, 2026 - https://compute.events/paris2026/cfp.html -- Guillaume Lemaitre

1 0

[ANN] scikit-learn 1.9.0rc1 is online ! Please test
by Jérémie du Boisberranger May 20, 2026

May 20, 2026

Hi everyone, Please help us test the first release candidate for scikit-learn 1.9: pip install scikit-learn==1.9.0rc1 Changelog: https://scikit-learn.org/dev/whats_new/v1.9.html This release features a few significant changes that requires special attention: - narwhals <https://narwhals-dev.github.io/narwhals/> was adopted as a new dependency to better handle dataframe interoperability. The minimum supported version is 2.0.1. - A new config <https://scikit-learn.org/stable/modules/generated/sklearn.set_config.html> option was added to control the sparse interface of all sparse objects that scikit-learn produces. Set `sklearn.set_config(sparse_interface="sparray")` to output sparse arrays instead of sparse matrices (current default). The default will change in a future release. - SLEP023: Callback API <https://scikit-learn-enhancement-proposals.readthedocs.io/en/latest/slep023…> has been accepted and this release ships a minimal implementation, with callback support in LogisticRegression(solver="lbfgs"), Pipeline, StandardScaler and all *SearchCV classes. The callback API <https://scikit-learn.org/1.9/callbacks.html> is experimental for now and might change without deprecation. Please let us know about any regression, in particular if you maintain a project with a dependency on scikit-learn. Thanks to everyone who contributed to this release! Jérémie, on behalf of the scikit-learn maintainer team.

1 0

What happens to hyperparam optimisation when roc_auc ==1?
by Sole Galli May 13, 2026

May 13, 2026

Hello community! I am training several models with hyperparam optimisation. I am using successive halving but I think the question applies to any search algo. I am optimising for roc_auc_score. Some of the datasets are very imbalanced, which means that the roc can't be calculated and it is set to 1. Following error: UndefinedMetricWarning: No positive samples in y_true, true positive value should be meaningless I guess the obvious answer is that I should optimise for something else, like the balanced accuracy, but the search runs anyways, so I was wondering which hyperparameters are returned? the first tested? Thanks a lot! Best Sole

1 0

Narwhals as a dependency
by Adrin May 7, 2026

May 7, 2026

Hi, We've added narwhals as a dependency to handle dataframes in scikit-learn in #31127 <https://github.com/scikit-learn/scikit-learn/pull/31127>. It's now merged for us to test edge cases before the next release. If there are any major issues or objections, we can revert this before the release. This is thanks to everyone who contributed, especially Christian who authored the PR, as well as narwhals folks holding our hands through the process. Thanks, Adrin

2 1