[CyberNativeAI]

Link to issue

There is no separate issue: this PR carries the report and a reproduction. We can open an issue as well if you would rather track it there.

Description of change

px.ecdf counts a row whose value is missing (pandas NaN/None, nullable <NA>, polars or pyarrow null) as an observation, so the curve stops short of 1 and gets an extra point at the missing x. This PR drops those rows from each group before the cumulative sum.

import pandas as pd
import plotly.express as px

fig = px.ecdf(pd.DataFrame({"v": [3, 1, 2, None]}), x="v")
print(fig.data[0].x.tolist(), fig.data[0].y.tolist())

prints this on current main (7.1.0 gives the same values):

[1.0, 2.0, 3.0, nan] [0.25, 0.5, 0.75, 1.0]

The expected result is y = [1/3, 2/3, 1] at x = [1, 2, 3], with no point at nan. All three observed values are at or below 3, so by the docstring ("values represent data at or below the point") the curve should be 1 there, not 0.75.

The same happens with ecdfnorm="percent" (75 at the largest value), ecdfnorm=None (an extra point with count 4), ecdfmode="complementary" and "reversed", orientation="h", and within each color group. Data without missing values gives the same values as before.

Cause: in make_figure, the ecdfmode block sorts each group with nulls_last=True, then takes cum_sum() and the group total over every row, null rows included. The fix filters out rows where the base column is null just before that sort. The pandas code in 5.24.1 had the same logic, so this is not new.

Demo

Numbers instead of a screenshot, from the snippet above: before [1.0, 2.0, 3.0, nan] [0.25, 0.5, 0.75, 1.0], after [1.0, 2.0, 3.0] [0.3333333333333333, 0.6666666666666666, 1.0].

Testing strategy

Adds test_ecdf_ignores_missing_values to tests/test_optional/test_px/test_px_functions.py, parametrized over the suite's dataframe constructors and the three ecdfmode values (15 cases). On main (fcc3c8e) without the fix, those 15 fail and everything else in tests/test_optional/test_px passes; with the fix, all 1081 tests there pass (10 skipped, 4 xfailed). We ran only tests/test_optional/test_px, and test_imshow.py and test_trendline.py were not collected because xarray and pytz were not installed in our environment.

Guidelines

• I have reviewed the pull request guidelines and the Code of Conduct and confirm that this PR follows them.
• I have added an entry to the changelog if needed (not required for documentation PRs).

Prepared by CyberNative AI LLC, an AI-run company, using AI agents.