✔️ Linting Passed

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

_{Generated for commit: 7f7e30e. Link to the linter CI: here}

@lucyleeow could you please maybe have a look?

Could you update the ROC curve display examples to set %matplotlib widget and see if it renders as expected in the HTML doc?

Agreed that it would be nice to see how this works inside a notebook with %matplotlib widget (need to install ipympl I think). Maybe I am wrong but I don't think there is a way for sphinx-gallery to capture the interactive widget so I am not sure it will work with our HTML doc.

Could you update the ROC curve display examples to set %matplotlib widget and see if it renders as expected in the HTML doc?

How can I add %matplotlib widget in an example and still make it run as a python script ?

Maybe the easiest way to try it yourself is to fetch my branch and run one of these:

• single curve

from sklearn.datasets import make_classification
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import RocCurveDisplay
import matplotlib.pyplot as plt
 
X, y = make_classification(n_samples=200, class_sep=0.5)
clf = LogisticRegression().fit(X, y)
RocCurveDisplay.from_estimator(clf, X, y)
plt.show()

• cv results

import matplotlib.pyplot as plt
from sklearn.datasets import make_classification
from sklearn.metrics import RocCurveDisplay
from sklearn.model_selection import cross_validate
from sklearn.linear_model import LogisticRegression

X, y = make_classification(n_samples=200, class_sep=0.5, random_state=0)
clf = LogisticRegression()
cv_results = cross_validate(
    clf, X, y, cv=3, return_estimator=True, return_indices=True
)
RocCurveDisplay.from_cv_results(cv_results, X, y)
plt.show()

• several displays

from sklearn.metrics import RocCurveDisplay
from sklearn.datasets import make_classification
from sklearn.linear_model import LogisticRegression
import matplotlib.pyplot as plt

X, y = make_classification(n_samples=200, class_sep=0.5)
X_, y_ = make_classification(n_samples=200, class_sep=0.85)
clf = LogisticRegression().fit(X, y)
clf2 = LogisticRegression().fit(X_, y_)
disp = RocCurveDisplay.from_estimator(clf, X, y)
RocCurveDisplay.from_estimator(clf2, X_, y_, ax=disp.ax_)
plt.show()

How can I add %matplotlib widget in an example and still make it run as a python script?

That's a good question, I spoke too quickly because I am used to working with jupytext/jupyterbook on other projects. For the scikit-learn doc, we would probably need non-trivial changes sphinx-gallery to allow for this.

I know that jupytext for instance can convert python script with commented out lines that start with # %magic command into a notebook with the uncommented %magic command in the generated .ipynb file before rendering it with jupyter. Something similar would have to be implemented in sphinx gallery.

BTW, I tried in a VS Code interactive window with %matplotlib widget after installing the ipympl package, and it works great.

The mixin is in _plotting and uses generic variable names because it's not dedicated to RocCurveDisplay, but is instead ready to be used in other displays.

Can you try to add it to at least one other display (e.g. PR curve) to check that the design is generic enough to be reused?

Can you try to add it to at least one other display (e.g. PR curve) to check that the design is generic enough to be reused?

Done in e916a77.

That's a good question, I spoke too quickly because I am used to working with jupytext/jupyterbook on other projects. For the scikit-learn doc, we would probably need non-trivial changes sphinx-gallery to allow for this.

I know that jupytext for instance can convert python script with commented out lines that start with # %magic command into a notebook with the uncommented %magic command in the generated .ipynb file before rendering it with jupyter. Something similar would have to be implemented in sphinx gallery.

Looks like there's a way but in the SG config file (see d0735d5). It's not visible in the HTML rendered examples but is there in the generated notebook. In all generated notebooks actually.

I wanted to check the jupyterlite button from the generated example but the tooltip doesn't show up :(

Also I didn't find where to request ipympl to be installed for jupyterlite. @lesteve can you show me ?

By the way, sphing-gallery supports examples using plotly (see https://sphinx-gallery.github.io/dev/auto_plotly_examples/plot_0_plotly.html for instance) so it might be possible to have support for matplotlib interactive plots at some point, we'll see.

What would be required for an interactive plot to be rendered - i.e. what would the rst look like?

Plotly simply uses the capture_repr configuration - essentially the following rst header:

.. raw:: html

        <div class="output_subarea output_html rendered_html output_result">

followed by the html of the figure.

Note that matplotlib animations are also supported. Depending on what is set as the format in config, either raw html is embedded (similar to above) or a video file is saved and a video directive (I think this needs the sphinxcontrib-video extension) is used.
See: https://github.com/sphinx-gallery/sphinx-gallery/blob/df6ca880069a891b565fb62598a036f21d55ed27/sphinx_gallery/scrapers.py#L233