DOC improve docs in DecisionBoundaryDisplay and linear models #30729

StefanieSenger · 2025-01-28T12:56:09Z

This PR proposes doc improvements in the docstring of DecisionBoundaryDisplay.from_estimator and in and linear_models.rst to facilitate readability.

github-actions · 2025-01-28T12:57:40Z

✔️ Linting Passed

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

_{Generated for commit: 2227ef9. Link to the linter CI: here}

StefanieSenger · 2025-01-28T12:57:40Z

sklearn/inspection/_plot/decision_boundary.py

@@ -248,14 +248,12 @@ def from_estimator(
            :func:`contour <matplotlib.pyplot.contour>`,
            :func:`pcolormesh <matplotlib.pyplot.pcolormesh>`.

-        response_method : {'auto', 'predict_proba', 'decision_function', \
+        response_method : {'auto', 'decision_function', 'predict_proba', \


Mentioning the options for response_method in the same order as they are checked through to reduce mental load for users.

Thanks for this fix.

I find it weird that we don't attempt to use predict_proba first when available.

Maybe we could open a dedicated follow-up PR if people agree this would be a better default.

It seems to me that predict_proba is less custom and easier to interpret than decision_function, so yes if users use the default auto they should maybe better see the more general info. Also, it would be good to add a title to the plot informing the user which response_method was used.

StefanieSenger · 2025-01-28T12:59:31Z

doc/modules/linear_model.rst

+:class:`LinearRegression` will take in its ``fit`` method arguments ``X``, ``y``,
+``sample_weight`` and will store the coefficients :math:`w` of the linear model in its
+``coef_`` and ``intercept_`` attributes::


Mentioning intercept_ attribute to avoid misunderstanding as if LinearRegression fits intercept by default.

I also substituted arrays with arguments, since the input can be array-likes.

We should also drop the "will" and instead use the present tense while we are at it.

I have made it present tense.

StefanieSenger · 2025-01-28T16:06:48Z

There seems to be a floating point rounding issue with some CIs expecting np.float64(1.1102230246251565e-16) and others np.float64(2.22044...e-16) (twice the first number). These are both close to 0 and I am trying to find out how to test just near closeness to zero.

Otherwise, we could change the doc example into

>>> np.isclose(reg.intercept_, 0)
True

maybe?

StefanieSenger · 2025-01-28T18:22:02Z

I have solved it by rounding. Please let me know if I should do it another way.

ogrisel · 2025-01-29T16:27:13Z

doc/modules/linear_model.rst

+:class:`LinearRegression` will take in its ``fit`` method arguments ``X``, ``y``,
+``sample_weight`` and will store the coefficients :math:`w` of the linear model in its
+``coef_`` and ``intercept_`` attributes::


We should also drop the "will" and instead use the present tense while we are at it.

doc/modules/linear_model.rst

ogrisel · 2025-01-29T16:31:11Z

sklearn/inspection/_plot/decision_boundary.py

@@ -248,14 +248,12 @@ def from_estimator(
            :func:`contour <matplotlib.pyplot.contour>`,
            :func:`pcolormesh <matplotlib.pyplot.pcolormesh>`.

-        response_method : {'auto', 'predict_proba', 'decision_function', \
+        response_method : {'auto', 'decision_function', 'predict_proba', \


Thanks for this fix.

I find it weird that we don't attempt to use predict_proba first when available.

Maybe we could open a dedicated follow-up PR if people agree this would be a better default.

Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>

betatim · 2025-01-30T14:46:29Z

Merged because it looked good, not to end the discussion :)

DeaMariaLeon · 2025-02-04T14:45:30Z

Hi @StefanieSenger, I get this on my mac:

_______________________________ [doctest] linear_model.rst ________________________________
041 ``sample_weight`` and stores the coefficients :math:`w` of the linear model in its
042 ``coef_`` and ``intercept_`` attributes::
043 
044     >>> from sklearn import linear_model
045     >>> reg = linear_model.LinearRegression()
046     >>> reg.fit([[0, 0], [1, 1], [2, 2]], [0, 1, 2])
047     LinearRegression()
048     >>> reg.coef_
049     array([0.5, 0.5])
050     >>> reg.intercept_
Expected:
    0.0
Got:
    np.float64(4.440892098500626e-16)

/Users/dealeon/projects/scikit-learn/doc/modules/linear_model.rst:50: DocTestFailure
================================= short test summary info =================================
FAILED linear_model.rst::linear_model.rst

What do you think? It's not blocking me in any way, it's just FYI.. or if I should do something about it (open an issue or so). Thanks

StefanieSenger · 2025-02-04T14:55:06Z

Hi @DeaMariaLeon,

thanks for bringing this up. I also get this locally with python -m doctest doc/modules/linear_model.rst, see #30729 (comment).

I would think this might be due to differences between by local doctest setup and the CI doctest setup, but not sure to be honest.

A lot of other doctest things are also failling for me because of different rounding. It would be nice to learn how to mimic the CI's doctest locally, because it could save time in the future.

DOC improve docs in DecisionBoundaryDisplay and linear models

1a04248

github-actions bot added module:inspection Documentation labels Jan 28, 2025

StefanieSenger commented Jan 28, 2025

View reviewed changes

attempt of fixing doctest

e9a61f1

attempt of fixing doctest

d6b9193

ogrisel approved these changes Jan 29, 2025

View reviewed changes

StefanieSenger and others added 2 commits January 30, 2025 10:00

apply review

52f9e1c

Update doc/modules/linear_model.rst

2227ef9

Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>

betatim approved these changes Jan 30, 2025

View reviewed changes

betatim merged commit a996f43 into scikit-learn:main Jan 30, 2025
31 checks passed

StefanieSenger deleted the doc_improvements branch January 30, 2025 15:40

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

DOC improve docs in DecisionBoundaryDisplay and linear models #30729

DOC improve docs in DecisionBoundaryDisplay and linear models #30729

StefanieSenger commented Jan 28, 2025

github-actions bot commented Jan 28, 2025 •

edited

Loading

StefanieSenger Jan 28, 2025

ogrisel Jan 29, 2025

StefanieSenger Jan 30, 2025

StefanieSenger Jan 28, 2025

ogrisel Jan 29, 2025

StefanieSenger Jan 30, 2025

StefanieSenger commented Jan 28, 2025 •

edited

Loading

StefanieSenger commented Jan 28, 2025

ogrisel Jan 29, 2025

ogrisel Jan 29, 2025

betatim commented Jan 30, 2025

DeaMariaLeon commented Feb 4, 2025

StefanieSenger commented Feb 4, 2025 •

edited

Loading

DOC improve docs in DecisionBoundaryDisplay and linear models #30729

DOC improve docs in DecisionBoundaryDisplay and linear models #30729

Conversation

StefanieSenger commented Jan 28, 2025

github-actions bot commented Jan 28, 2025 • edited Loading

✔️ Linting Passed

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

StefanieSenger commented Jan 28, 2025 • edited Loading

StefanieSenger commented Jan 28, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

betatim commented Jan 30, 2025

DeaMariaLeon commented Feb 4, 2025

StefanieSenger commented Feb 4, 2025 • edited Loading

github-actions bot commented Jan 28, 2025 •

edited

Loading

StefanieSenger commented Jan 28, 2025 •

edited

Loading

StefanieSenger commented Feb 4, 2025 •

edited

Loading