Hello @benHeid,

thanks for reporting.

I have checked and would also think it is a bug. This code returned False up until version 1.5.2 and since version 1.6 it does return True:

from sklearn.preprocessing import StandardScaler, MinMaxScaler
print(StandardScaler()._get_tags()["stateless"])
print(MinMaxScaler()._get_tags()["stateless"])

As a workaround please use the "requires_fit" tag, which is supposed to replace the "stateless" tag.

The issue is related to #30327.

As a workaround please use the "requires_fit" tag

Actually it is the right way to do with the new tag infrastructure

from sklearn.utils import get_tags
from sklearn.preprocessing import StandardScaler

get_tags(StandardScaler()).requires_fit

And indeed, there is a bug with the conversion if the old tag infrastructure that we need to solve.

So the bug is here:

https://github.com/scikit-learn/scikit-learn/blob/main/sklearn/utils/_tags.py#L590

We should change to:

        "stateless": not new_tags.requires_fit,

@EmilyXinyi, would you like to take care of that? Don't feel obliged though, only if you like.

Hi @StefanieSenger thanks for flagging! I can probably take care of this next weekend, but if it's an urgent change that needs to go out before then, I most likely won't have time during the work week.
(I will come back to this issue next weekend to check for updates and possible put in a fix)

I just put in a PR for a fix, but out of curiosity, what is the purpose of this tag thing? I have not used this functionality before, I can't seem to understand it very well just by reading the associated code, so if someone could please enlighten me that would be greatly appreciated 😸

Maybe the documentation can help: https://scikit-learn.org/stable/developers/develop.html#estimator-tags

In short, the first use of it was to give some information regarding the capabilities of an estimator and it was used in the common tests to know whether an estimator is compatible with the scikit-learn API.

Here, the stateless tag enforce a specific behaviour of the estimator that is specifically tested in the common test: an estimator can be used by calling transform without a previous call to fit.

Nowadays, we sometime use the tag in the source code as well (not only in test).

Closing since we end the deprecation cycle and it will be fixed in the up coming 1.7 release.