✔️ Linting Passed

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

_{Generated for commit: 3094eca. Link to the linter CI: here}

I think I'm fine with the name. I was looking if can have the word counts in the name of the function but it starts to be really long. So I would be OK with the proposed name.

@adrinjalali There is an issue with numpydoc validation in the binary_classification_curve function, the error is RT03: Return value has no description. Do you know why this is? I already documented the return value of the function.

When you look at the CI log, this is the error, not the return:

[gw0] linux -- Python 3.12.7 /usr/share/miniconda/envs/testvenv/bin/python
810         Decreasing score values.
811 
812     Examples
813     -------
814     >>> import numpy as np
815     >>> from sklearn.metrics import binary_classification_curve
816     >>> y_true = np.array([0, 0, 1, 1])
817     >>> y_scores = np.array([0.1, 0.4, 0.35, 0.8])
818     >>> fps, tps, thresholds = binary_classification_curve(y_true, y_scores)
819     >>> fps
Expected:
    array([0, 1, 1, 2])
Got:
    array([0., 1., 1., 2.])

You just need to fix the output to floats

You just need to fix the output to floats

I fixed this one. But this is the log I was talking about

E           array([0.8, 0.4, 0.35, 0.1])
E           
E           # Errors
E           
E            - RT03: Return value has no description
E            - RT03: Return value has no description
E            - RT03: Return value has no description
E            - RT03: Return value has no description
E            - RT03: Return value has no description
E            - RT03: Return value has no description
E            - RT03: Return value has no description
E            - RT03: Return value has no description
E            - RT03: Return value has no description
E            - RT03: Return value has no description
E            - RT03: Return value has no description
E            - RT03: Return value has no description
E            - RT03: Return value has no description

function_name = 'sklearn.metrics._ranking.binary_classification_curve'
msg        = '\n\n/home/vsts/work/1/s/sklearn/metrics/_ranking.py\n\nTested function: sklearn.metrics._ranking.binary_classificatio...3: Return value has no description\n - RT03: Return value has no description\n - RT03: Return value has no description'
request    = <FixtureRequest for <Function test_function_docstring[sklearn.metrics._ranking.binary_classification_curve]>>
res        = {'deprecated': False, 'docstring': 'Calculate true and false positives per binary classification threshold.\n\nParamet...n'), ('RT03', 'Return value has no description'), ...], 'file': '/home/vsts/work/1/s/sklearn/metrics/_ranking.py', ...}

This test tests.test_docstrings.py::test_function_docstring[sklearn.metrics._ranking.binary_classification_curve] is failing.

That's indeed very odd. @StefanieSenger @lucyleeow would you have an idea here?

By looking at the example change, I went back to the issue to understand the real use case here. So it seems that users are interested to get the entry of the confusion matrix to be able to compute different metrics of interest.

I'm wondering that what we have here is not enough since we only return the TP and FP and we don't provide any information about the negatives. Eg. computing the two samples Kolmogorov-Smirnov test (https://en.wikipedia.org/wiki/Kolmogorov%E2%80%93Smirnov_test) could require these data.

@adrinjalali would you think that if we make this function public, we should also return the TN and FN for completness?

Yeah I think that makes sense @glemaitre .

Also related, we probably want to get #15522 merged

@glemaitre it is possible to compute the negatives from the positives. Since the tps array is increasing, false negatives are given by tps[-1] - tps. Similarly, true negatives are given by fps[-1] - fps. _binary_clf_curve does not return negatives because they are easy to calculate.

@SuccessMoses You will need an additional statistic most probably to infer the four statistics. However, on a user perspective, it is really annoying to have to carry on some code to compute the negative part. I would like just to call the function, get my counts and then deal with them. It is a nicer user experience.