ENH: Display parameters in HTML representation #30763

DeaMariaLeon · 2025-02-03T15:28:26Z

Reference Issues/PRs

Working on first point on #26595
Fixes: #21266

What does this implement/fix? Explain your changes.

This code allows to visualise the estimator's parameter values. It adds an interactive dropdown hidden by default.
Copy paste button added. When clicking on it, the parameter's name is saved to the clipboard.

Any other comments?

For my future-self: See https://ipython.readthedocs.io/en/stable/config/integrating.html for information on _repr_mimebundle_

@glemaitre : WDYT?

github-actions · 2025-02-03T15:29:52Z

✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 3d5804f. Link to the linter CI: here}

betatim · 2025-02-07T15:56:27Z

This looks super cool! What about a more descriptive title for the PR? Every time I see a notification related to it pop up I think: what get_params is broken??!! To one second later realise what this PR is really about :D

Maybe something like "Display parameters in HTML representation"?

trallard · 2025-05-13T11:48:10Z

Hey folks, I am so happy to see more accessibility discussions in the wild ✨💜 Thank you for being thoughtful and striving to serve our community better.
There was also another talk at PyCon DE this year by Smera, our fabulous designer!

If you wanted to make the style more in line with the PyData Sphinx Theme:

We have our Design System documented at https://pydata-sphinx-theme.readthedocs.io/en/stable/community/design-system.html
All our components and specifications are also in a community Figma file https://www.figma.com/community/file/1443191723065200671

The reason I bring these resources is per @betatim 's comment:

There was a great talk at PyCon Germany about the work that went into pydata-sphinx-theme to make it more accessible and one of my main takeaways from it was that accessibility is a bit like cryptography: so many things to think about that it is worth not rolling your own :D

We have spent much time thinking about accessibility and design for PyData Sphinx Theme and documenting our guidelines so our community can reuse and build on them without reinventing the wheel.
So, you might find some of our guidelines and styles helpful for you moving forward.
Alternatively, if you need or would like accessibility-related feedback or brainstorming, feel free to contact me, and I will be more than happy to help.

DeaMariaLeon · 2025-05-13T14:45:51Z

Here is the example again: Column Transformer with Mixed Types

ArturoAmorQ · 2025-05-13T15:54:44Z

Sorry for jumping in. I don't want to bring noise to a PR that already has 6 participants. I just wanted to share that I had a look on the render linked in #30763 (comment) and I noticed a change in behavior.

On main:
This PR:

Personally I find it more useful to know the column names than knowing they were selected using a make_column_selector object. Is there a way to retrieve the actual names of the selected columns from the object?

DeaMariaLeon · 2025-05-13T16:16:54Z

@ArturoAmorQ thanks for bringing it up!.. it's a bug..
The names of the columns are shown correctly before make_column_selector is used (the first diagram in the document is fine).

glemaitre · 2025-05-13T16:23:56Z

I don't think that there is a bug here. @ArturoAmorQ I think that you click on the first pipeline in main and the second pipeline in this PR. In main, it is the rendering of the the second pipeline:

I was a bit surprise because the code never execute any function when it comes to diagram. However, it could be a further improvement potentially by recording the column name once fit is called and show them once the pipeline is fitted. Before fit, we cannot do better than the signature of the object. We could also the representation of the make_column_selector function to have a better string reprenstation.

ArturoAmorQ · 2025-05-13T16:26:19Z

Actually it's me who read the example too fast. Sorry for that. There are 2 pipelines in the example and the second one uses the make_column_selector. Your PR has the same behavior as before.

Still I wonder if we can somehow retrieve the column names from the make_column_selector 🤔 In any case that would be for a different PR.

Sorry again for the noise.

glemaitre · 2025-05-13T16:31:53Z

Thanks @trallard for the guideline. We definitely need some visual improvements with those diagrams. Basically, this PR was a good starting point to find out a couple of limitations: inconsistency between IDEs/doc rendering and dark/light theme management (among others).

We should probably limit the scope of this PR but we need to revisit specifically the look of those diagrams and those guidelines will be useful for sure. We can hopefully also reuse knowledge that we gained when developing a couple of add-ons in skrub.

trallard · 2025-05-13T17:43:49Z

+1 on limiting scope.
If there is anything I can help with just give me a ping whenever.

GaelVaroquaux · 2025-05-13T19:41:44Z

@trallard , happy to have you around. Thanks a lot!

GaelVaroquaux

The rendering looks fine to me (I no longer have issues with the contrast).

I do have one question on the code (inlined in the review).

In addition, I just wanted to check: this refactor does not change the protocol for subclasses in other packages compared to the one that we had before? If it does, we should document this clearly.

GaelVaroquaux · 2025-05-20T11:21:12Z

sklearn/utils/_repr_html/base.py

+        return self._repr_html_inner
+
+    def _repr_html_inner(self):
+        return self._html_repr()


Why do we need this method? It seems to me that we could simply call _html_repr istead of this

This was taken from the original base.py code (that's the way it is in main). But I forgot to add its docstring:

This function is returned by the @property `_repr_html_` to make `hasattr(estimator, "_repr_html_") return `True` or `False` depending on `get_config()["display"]`.

If I should do it differently of course I'll do it, but in the meantime I added the docstring.

DeaMariaLeon · 2025-05-20T14:25:43Z

this refactor does not change the protocol for subclasses in other packages compared to the one that we had before?

No, it doesn't.

glemaitre

Thanks @GaelVaroquaux for the review. Looks good on my side. Let's merge.

Thanks @DeaMariaLeon it is a nice improvement.

DeaMariaLeon · 2025-05-21T08:36:13Z

Thank you All!

GaelVaroquaux · 2025-05-21T08:37:08Z

Great work! Thanks a lot!!

betatim · 2025-05-21T08:39:59Z

Whoop whoop 🥳

GaelVaroquaux · 2025-05-21T08:42:49Z

I think that this should be a highlight of the release, by the way

betatim · 2025-05-21T11:15:23Z

I tried this out (for the first time) just now. One thing I noticed is that there is a "copy" button for each parameter. Not having paid attention to the discussion.design of this I approached it like an uneducated user. My first thought was that maybe it will copy the argument and value (n_neighbors=20) but it copies only the parameter name (n_neighbors). Overall I was wondering what the idea behind the feature was?

glemaitre · 2025-05-21T12:43:08Z

@betatim The feature is particularly useful when you have nested pipeline and column transformer to get the fully qualified name, e.g. histgradientboostinclasssifier__max_depth. With a single estimator, it is a bit less pertinent. You can see it as an alternative to call pipeline.get_params to find the fully qualified name to copy it and paste it in pipeline.set_params (e.g. when doing the parameter grid in GridSearchCV.

We started only by copying the parameter name but I don't mind extending to the parameter value if we think it is useful.

betatim · 2025-05-21T13:11:11Z

That makes sense. Maybe something to show off in the release highlight to help others realise why copying the name is useful

Co-authored-by: Guillaume Lemaitre <guillaume@probabl.ai>

DeaMariaLeon added 9 commits January 31, 2025 09:01

first try

f1540bd

wip

d4e6359

wip

53e26d6

wip

27a2934

wip

1363a90

wip

1f44e8d

wip

ae18394

wip

bf1774b

wip

fbddf12

github-actions bot added the module:utils label Feb 3, 2025

DeaMariaLeon changed the title ~~WIP: get_params() work - not ready for review~~ WIP: get_params() work Feb 3, 2025

DeaMariaLeon added 5 commits February 3, 2025 18:15

changed button color

3cedcf4

Merge remote-tracking branch 'upstream/main' into get-params

65a6fe9

removed svg - use fontawesome

2e7964d

wip

4fff750

wip

5da5276

glemaitre self-requested a review February 6, 2025 15:28

DeaMariaLeon added 3 commits February 6, 2025 18:04

single estimator - broken pipe

719871f

fixed pipeline - I think

6fb104b

changing size colors

ea83b49

removed name

4de9ee0

DeaMariaLeon changed the title ~~WIP: get_params() work~~ WIP: Display parameters in HTML representation Feb 7, 2025

DeaMariaLeon added 6 commits February 8, 2025 12:45

script at the bottom

79c5133

added temporary Copied message

0f7fae9

removed console log

2bcf71a

correct margin

9f580f7

found non-defaults params

38bc5b5

Font color black when non-default

6e4c191

DeaMariaLeon added 2 commits May 13, 2025 12:07

Merge remote-tracking branch 'upstream/main' into get-params

2110981

[doc build] trigger doc

736bc6b

[doc build] trigger doc and Merge remote-tracking branch 'upstream/main'

f0e7224

GaelVaroquaux approved these changes May 20, 2025

View reviewed changes

DeaMariaLeon added 2 commits May 20, 2025 15:59

Added docstring to _repr_html_inner

e8dd6b8

Merge remote-tracking branch 'upstream/main' into get-params

3d5804f

glemaitre approved these changes May 21, 2025

View reviewed changes

glemaitre merged commit d077f82 into scikit-learn:main May 21, 2025
36 checks passed

DeaMariaLeon deleted the get-params branch May 21, 2025 08:36

DeaMariaLeon mentioned this pull request May 28, 2025

ENH Display fitted attributes in HTML representation #31442

Open

jeremiedbb pushed a commit to jeremiedbb/scikit-learn that referenced this pull request May 30, 2025

ENH: Display parameters in HTML representation (scikit-learn#30763)

94ababc

Co-authored-by: Guillaume Lemaitre <guillaume@probabl.ai>

jeremiedbb pushed a commit that referenced this pull request Jun 5, 2025

ENH: Display parameters in HTML representation (#30763)

a934685

Co-authored-by: Guillaume Lemaitre <guillaume@probabl.ai>

glemaitre mentioned this pull request Jun 6, 2025

FIX short deprecation cycle for private module #31500

Open

Uh oh!

ENH: Display parameters in HTML representation #30763

ENH: Display parameters in HTML representation #30763

Uh oh!

Conversation

DeaMariaLeon commented Feb 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Reference Issues/PRs

What does this implement/fix? Explain your changes.

Any other comments?

Uh oh!

github-actions bot commented Feb 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✔️ Linting Passed

Uh oh!

betatim commented Feb 7, 2025

Uh oh!

trallard commented May 13, 2025

Uh oh!

DeaMariaLeon commented May 13, 2025

Uh oh!

ArturoAmorQ commented May 13, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

DeaMariaLeon commented May 13, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

glemaitre commented May 13, 2025

Uh oh!

ArturoAmorQ commented May 13, 2025

Uh oh!

glemaitre commented May 13, 2025

Uh oh!

trallard commented May 13, 2025

Uh oh!

GaelVaroquaux commented May 13, 2025

Uh oh!

GaelVaroquaux left a comment

Choose a reason for hiding this comment

Uh oh!

GaelVaroquaux May 20, 2025

Choose a reason for hiding this comment

Uh oh!

DeaMariaLeon May 20, 2025

Choose a reason for hiding this comment

Uh oh!

DeaMariaLeon May 20, 2025

Choose a reason for hiding this comment

Uh oh!

DeaMariaLeon commented May 20, 2025

Uh oh!

glemaitre left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

DeaMariaLeon commented May 21, 2025

Uh oh!

GaelVaroquaux commented May 21, 2025 via email

Uh oh!

betatim commented May 21, 2025

Uh oh!

GaelVaroquaux commented May 21, 2025 via email

Uh oh!

betatim commented May 21, 2025

Uh oh!

glemaitre commented May 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

betatim commented May 21, 2025

Uh oh!

Uh oh!

DeaMariaLeon commented Feb 3, 2025 •

edited

Loading

github-actions bot commented Feb 3, 2025 •

edited

Loading

ArturoAmorQ commented May 13, 2025 •

edited

Loading

DeaMariaLeon commented May 13, 2025 •

edited

Loading

glemaitre commented May 21, 2025 •

edited

Loading