Christian Lorentzen

#### Reference Issues/PRs Fixes #20998. #### What does this implement/fix? Explain your changes. This PR adds argument `sparse` to `SplineTransformer`. Set to `True`, it returns a sparse csr matrix. ####...

### Describe the workflow you want to enable As B-splines naturally have a sparse structure, I'd like to have the option that `SplineTransformer` returns a sparse matrix instead of always...

This meta issue aims at giving an overview/summary of all quantile related issues and pull requests. This may help to coordinate future work. - [x] `metrics`: Add quantile loss as...

[MRG] DOC add guideline for choosing a scoring function

11

Solves issue #10584. Adds the new chapter "3.3.2. Which scoring function should I use?" to the user guide.

Formula for dual gap of elastic net in coordinate descent solver

13

### Describe the bug The computation of the dual gap for the elastic net in the coordinate descent solver (`enet_coordinate_descent`) might be wrong. The elastic net minimizes ``` Primal(w) =...

FEA add Tweedie log likelihood

1

- [x] ~~closes~~ addresses #7310 - [x] tests added / passed. - [ ] code/documentation is well formatted. - [x] properly formatted commit message. See [NumPy's guide](https://docs.scipy.org/doc/numpy-1.15.1/dev/gitwash/development_workflow.html#writing-the-commit-message). - [ ]...

Add Tweedie distribution: likelihood, pdf and sampling methods

16

#### Is your feature request related to a problem? Please describe I'd like to have the Tweedie distribution in order to: - calculate the likelihood - calculate the pdf -...

GLM fit without penalty

4

I would be very nice to also fit completely unpenalized GLMs (as in base R glm). For categorical features, one then needs to drop one level, called reference or base...

I think, the comments https://github.com/Quantco/glum/blame/main/src/glum/_cd_fast.pyx#L115_L116 need to be updated. I guess the definition of the argument `q` changed during some code optimization from `q = X^T y` to `q =...

I'm wondering if the initial guess of GLM coefficients could be made faster by using random matrix sketching in the first iteration of Newton-Raphons/IRLS in the spirit of: - [Blendenpik:...