Christian Lorentzen

So currently, there is no critical voice, but still time to raise one. Please do so. Let's see how contributors would prefer to add narwhals: - 🚀 add narwhals as...

I have a question about API: Wouldn't it be something completely new for users to specify some amount of memory (RAM) limit? I would prefer to have a much simpler,...

The origin of the "friedman_mse" is the paper "Greedy function approximation: A gradient boosting machine." by Jerome H. Friedman [doi:1013203451](https://doi.org/10.1214/AOS/1013203451) around Eq 35. He even mentions that this is the...

Just read (or re-read) "Greedy Function Approxiations" again, in particular section 4.6 (and 4.5 and the reference "Additive logistic regression"). The point is that both cases use squared error, but...

> Ok but if both uses squared error, then the splitting criterion should be "squarred_error" and "friedman_mse" has no reason to exist (and anyway it computes exactly the same thing...

@cakedev0 I think there is no difference of squared error and Friedman squared error: `GradientBoostingClassifier` implements 1. order gradient descent in function space according to [1]. There are no weights...

> Although, note that HGB does line search for loss functions that aren't differentiable: see this [function/docstring](https://github.com/scikit-learn/scikit-learn/blob/c7d040e4f23e7888125de0af52e640329c8b9a5a/sklearn/ensemble/_hist_gradient_boosting/gradient_boosting.py#L74-L90) in the code. Hint: look at git blame😉 The author thought about future...

In general, good to merge those 2.

To be honest, I did not like #32197 too much and I did not review it. Now that we have it, the right thing to do is to set `excluded_set`...

> You mean outside this screening function? yes, right after its initialization.