Christian Lorentzen

In [Series evaluation of Tweedie exponential dispersion model densities](http://eprints.usq.edu.au/2335/1/Dunn_Smyth_Stats_and_Comp_v15n4.pdf) Eq.(1-2): ``` pdf(y, mu, p, phi) = f(y, theta, phi) = a(y, phi) exp(1/phi (y theta - kappa(theta))) kappa = cumulant...

@josef-pkt Where would you like to place this? Under `statsmodels.distributions` in a separate file? Which API to follow? An own class that quacks (duck-typing) almost like `scipy.stats.rv_continuous`?

It might be of interest that scikit-learn has a [`SplineTransformer`](https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.SplineTransformer.html#sklearn.preprocessing.SplineTransformer) which uses `BSpline` from scipy and has several knot selection methods, e.g. manual, uniform or quantile based. Using every data...

> So I'd really like to see an overall approach of how we'd like to converge to a recommended high-level API The `interpolate` module reminds me a lot of scikit-learn....

AFAIK, R also drops the first level so this seems like a good default, for sure, only in case of `alpha=0`. If someone still needs more control, she/he could go...

Let us discuss in #25572.

@scikit-learn/core-devs We need a decision here. Even a straight no is better then lingering. And with a bit of courage, we could get a little lift of the tree module...

> Polars' categorical type only maps to string categories. It would help me a lot to know, what I can do as a user. I read the error message and...

How can we make progress in finding a solution? Discuss it in a dev meeting, drafting a design document, just starting a PR, ...? I seems like a broad consensus...

I know of 3 reasons: - Pretty much any statical interpretation of weights requires them to be non-negative, e.g. aggregated observations/frequency weights define a weighted CDF and a CDF must...