# April 2021 26th -- Scikit-learn dev meeting
*Don't forget to sign in when editing and put your name in front of the entries you want to discuss*

### Need decision
- (Thomas) [#19914](https://github.com/scikit-learn/scikit-learn/pull/19914) GAM with Gradient boosted trees
- Here is an [example of the GAM](https://137869-843222-gh.circle-artifacts.com/0/doc/auto_examples/ensemble/plot_gam_boosted_regression_simple.html) in action!
- **Does this meet our inclusion criterion?**
- Extension: Explainable Boosting Machine -> https://arxiv.org/pdf/1909.09223.pdf
- Borderline with regards to inclusion criteron
- inclusion vs. external `scikit-learn-extra`?
- If there's a nice convincing example on the interpretability side, it'd be easier to decide.
- Thomas will do benchmarks
- (Guillaume) [#19880](https://github.com/scikit-learn/scikit-learn/issues/19880) Shape of predictions of multilabel classification
- Historical difficulty comes from the fact that OnevsRest and MultiLabel are built on the same code
- Look at the scorer / metrics to evaluate the RFC
- Start by making the reporting working with different output formats?
- (Guillaume) [#19881](https://github.com/scikit-learn/scikit-learn/issues/19881) Solving issue with `if_delegate_has_method` for nested meta-estimators
- At fit time, the current implementation fails when the base estimator is a meta estimator
- Solution: fitted estimator (or list of fitted estimators)
- Joel has a another approach PR with a callable implemenation with `available_if`: [#19948](https://github.com/scikit-learn/scikit-learn/pull/19948)
- Guillaume and Joel will consider the options and come back

### Need attention (review)
- (Guillaume) [#19963](https://github.com/scikit-learn/scikit-learn/issues/19963) Issue with MacOS/Python 3.9/clang flags
- Merged
- (Guillaume) [#19939](https://github.com/scikit-learn/scikit-learn/pull/19939) [#19952](https://github.com/scikit-learn/scikit-learn/pull/19952) GaussianProcess: issue consistency covariance/std. dev.
- Olivier will look
- (Guillaume) [#19982](https://github.com/scikit-learn/scikit-learn/pull/19982) Regression `DictVectorizer` with dense input and list of strings
- Easy review
- (Christian) [#19976](https://github.com/scikit-learn/scikit-learn/pull/19976) Remove deprecation of `loss="squared_error"` in HGBT and make it a hard renaming instead?
- 4 people favor being nice, and 2 people sticking for following experimental (all the rest did not review)
- (Christian) [#9978](https://github.com/scikit-learn/scikit-learn/pull/9978) Linear quantile regression
- Needs review
- (Christian) [#19263](https://github.com/scikit-learn/scikit-learn/pull/19263) ENH Makes ColumnTransformer more flexible by only checking for non-dropped columns
- Needs review for 1.0

### General topics

- (Guillaume) Bug fix release 0.24.2: [branching](https://github.com/scikit-learn/scikit-learn/pull/19954) preparing the release. Any major things that have been left-out?
- People need to help Guillaume
- (Chiara) https://draft.scientific-python.org/ scikit-learn participation in the Steering Council.
- Stefan van der Walt and Jarrod Millmann want to coordinate the scientific Python world (CI, common version numbers)
- (Guillaume) https://github.com/scipy/scipy/pull/13831 -> Back-compatibility for `pinv2` in scipy
- Thomas goes on scipy
- (Christian) How about an issue to award ideas and PRs that reduce the amount of code and enhance maintainability?
- (Adrin) NumFocus's D&I (Diversity & Inclusion) efforts and demographic survey efforts.
- Banner in documentation website
- (Adrin) NumFocus's education/training program/platform.
- It's an idea to have a platform where people/projects can offer training and use that as a way to financially support the project.
- (Adrin) **CoC:** We don't have a CoC council or email address.
- CoC email address used to file complaints
- CoC council handles complaints
- Needs to be addressed by creating a mailing list and council
- (Loïc) **MOOC:** Registrations to the "Machine learning in Python with scikit-learn" opened, see [this](https://www.fun-mooc.fr/en/courses/machine-learning-python-scikit-learn/). MOOC starts May 18. Most of the content (everything but videos and quiz solutions) is available [here](https://inria.github.io/scikit-learn-mooc/) and is CC-BY licensed.
- (Reshama) Data Umbrella Latin America Sprint

### Ongoing

- (Thomas) Working on rewriting [SLEP006 PR#55](https://github.com/scikit-learn/enhancement_proposals/pull/55). Does not need attention from everyone yet, but if you are curious here is a [rendered version of the rewrite](https://thomasjpfan.github.io/enhancement_proposals/slep006/proposal.html).
- (Joel) general issue seems to be that our technically best solution is also difficult to get one's head around.
- (Mathis and Julien) Started exploring pairwise distances improvements to have them run on chuncks on the two arrays with a reduction function to mitigate current bottlenecks in `KNeighborsMixin`'s
- https://github.com/mbatoul/sklearn_benchmarks
- Benchmark results and profiling reports: https://mbatoul.github.io/sklearn_benchmarks/
- [19950](https://github.com/scikit-learn/scikit-learn/pull/19950): Private Cython module proposal to make use of C++ std library for GIL-free sort-related reduction functions (to be used in k-nearest neighbors and kmeans++ for instance)
- [Some drafty notes one oneAPI's implementations](https://notes.inria.fr/uOCoCXIFRPKNo1v4j58E_A#)
- `OneHotEncoder` infrequent categories [#16018](https://github.com/scikit-learn/scikit-learn/pull/1601816018)
- Guillaume will review

### Next Meeting
- May 25th same time as today

### Contributors