BUG: fix "boolean value of NA is ambiguous" in iloc setitem with nullable Series RHS by IMGillusion · Pull Request #68312 · pandas-dev/pandas

IMGillusion · 2026-09-08T17:24:44Z

Issue and Reproducible Example

Assigning a nullable (e.g. Int64) Series containing a missing value as
the right-hand side of a Series.iloc setitem raised

TypeError: boolean value of NA is ambiguous

when the RHS had a non-default index:

import pandas as pd

ser = pd.Series([4, 6, 9, None, 10, 13, 15], index=[6, 1, 5, 0, 3, 2, 4], dtype="Int64")
indices = pd.Series([6, 1, 5, 0, 3, 2, 4], dtype="int64")
values = pd.Series([4, 6, 9, None, 10, 13, 15], index=[4, 1, 2, 6, 0, 5, 3], dtype="Int64")

ser.iloc[indices] = values

Root Cause

In _coerce_to_data_and_mask (pandas/core/arrays/numeric.py), the
precision-loss check compares the positionally-coerced float value against the
original:

idx = np.nanargmax(values)
if int(values[idx]) != original[idx]:

values is positionally aligned with the input, but when the input is a
Series with a non-default index, original[idx] performs a label lookup
and can land on a missing (NA) element even when the value at position idx
is concrete. Comparing a concrete value against NA yields NA, and the if
then raises "boolean value of NA is ambiguous".

Fix

Access the original value positionally (a Series is read via .iloc;
arrays and lists are already read positionally) and skip the check when the
original value is null. np.nanargmax already guarantees values[idx] is
concrete, so the positional original is not the missing element in practice; the
null guard is defensive.

Tests

test_setitem_iloc_rhs_series_non_default_index_with_na (new, in
pandas/tests/series/indexing/test_setitem.py) reproduces the issue and
asserts the outcome is positional: a non-default-indexed RHS yields the same
result as the equivalent default-indexed RHS.

This contribution was developed with the assistance of an AI coding assistant.

jbrockmendel · 2026-09-14T18:33:33Z

+            # is a Series with a non-default index, ``original[idx]`` would do
+            # a label lookup and can land on a missing element even when
+            # ``values[idx]`` is concrete (GH#62473). Access positionally.
+            original_idx = getattr(original, "iloc", original)[idx]


instead of getattr, can you use a more explicit isinstance

jbrockmendel · 2026-09-14T18:33:44Z

        else:
            idx = np.nanargmax(values)
-            if int(values[idx]) != original[idx]:
+            # `values` is positionally aligned with the input; when the input


Please ask your AI to make comments "maximally concise"

IMGillusion · 2026-09-14T19:41:49Z

Pushed 84c6bc5 to address both comments: the getattr is replaced with an explicit isinstance(original, ABCSeries) check, and the comment is trimmed to one line.

jbrockmendel · 2026-09-28T20:57:20Z

+            else:
+                original_idx = original[idx]
+            if (
+                not libmissing.checknull(original_idx)


dont we know we have floats here, so we can just use np.isnan?

actually, is this even reachable?

jbrockmendel · 2026-09-28T21:00:18Z

 - Bug in setitem (e.g. :meth:`Series.iloc`) silently storing wrong values when assigning a nullable or :class:`ArrowDtype` extension array into a NumPy-dtype column that could not hold it losslessly, such as a negative value into an unsigned-integer column (``-1`` became ``18446744073709551615``) or a value overflowing a narrower float dtype (:issue:`47776`)
 - Bug in setitem on nullable masked arrays (``Int64``, ``Float64``, ``boolean``, ...) silently coercing list-like values that the equivalent scalar setitem would reject — e.g. ``arr[[0]] = ['1']`` parsed the string into the numeric dtype, ``arr[[0]] = [True]`` on ``Int64``/``Float64`` coerced the boolean to ``1``, and an out-of-bounds value such as ``arr[[0]] = [1000]`` on ``Int8`` silently wrapped instead of raising (:issue:`45404`)
 - Fixed bug in :meth:`DataFrame.loc` where assigning an iterable to a single cell in an ``object`` dtype column incorrectly raised a ``ValueError`` (:issue:`26333`, :issue:`57962`)
+- Bug in :meth:`Series.iloc` setitem raising ``TypeError`` ("boolean value of NA is ambiguous") when the right-hand side is a nullable :class:`Series` (e.g. ``Int64``) containing a missing value with a non-default index, because the coercion compared the value positionally against the original by label (:issue:`62473`)


i suspect this also addresses cases without a missing value, e.g.

ser = pd.Series([1, 2], dtype="Int64") ser.iloc[[0, 1]] = pd.Series([1.0, 2.0], index=["a", "b"])

?

if so a test would be worthwhile too

IMGillusion · 2026-09-29T01:33:42Z

Addressed in ed45f9e. To answer the second question first: the checknull(original_idx) guard was unreachable, so I removed it rather than swap in np.isnan. idx = np.nanargmax(values) skips NaN slots, and values = np.asarray(original) keeps NaN exactly at original's NA positions — so the positionally-read original_idx is never NA and the guard could never fire. The positional read (original.iloc[idx]) is the actual fix; the guard was dead code.

IMGillusion · 2026-09-29T01:33:44Z

Yes, confirmed — and added the test. ser.iloc[[0,1]] = pd.Series([1.0, 2.0], index=['a','b']) raises KeyError on main with no missing value: same label-vs-position root cause. ed45f9e adds a no-NA regression test and broadens the whatsnew entry to cover both the TypeError (NA present) and KeyError (no NA) shapes.

github-actions · 2026-10-14T01:22:12Z

This pull request has had no activity from its author for 14 days, so I've marked it stale. If you've already addressed the feedback, re-request a review (the ↻ next to the reviewer) to move it back into the review queue. If you're still working on it, push a commit, reply to a review comment, or leave a comment — here or on the linked issue — and the Stale label clears itself. Otherwise it will be closed in 7 days to keep the queue manageable. Your branch still remains, and you can ask a maintainer to reopen this PR to continue. See the contributing guide for how the pull request lifecycle works.

Note: labels update via a once-a-day job, so Stale may take up to a day to clear.

IMGillusion · 2026-10-14T09:13:21Z

Thanks for the heads-up. The review feedback is fully addressed (84c6bc5: explicit isinstance check; ed45f9e: removed the unreachable checknull guard, added the no-NA regression test, broadened the whatsnew entry; 8502428: ruff format fix). CI is green on the current head. A re-review would be appreciated when there is time.

github-actions · 2026-10-29T01:10:06Z

This pull request has had no activity from its author for 14 days, so I've marked it stale. If you've already addressed the feedback, re-request a review (the ↻ next to the reviewer) to move it back into the review queue. If you're still working on it, push a commit, reply to a review comment, or leave a comment — here or on the linked issue — and the Stale label clears itself. Otherwise it will be closed in 7 days to keep the queue manageable. Your branch still remains, and you can ask a maintainer to reopen this PR to continue. See the contributing guide for how the pull request lifecycle works.

Note: labels update via a once-a-day job, so Stale may take up to a day to clear.

IMGillusion · 2026-10-29T05:39:07Z

Pinging again ahead of the stale auto-close. All of jbrockmendel's review feedback from 09-14 and 09-28 is addressed: isinstance check instead of getattr, the unreachable checknull guard removed, a no-NA regression test added, the whatsnew entry broadened to cover both the NA (TypeError) and no-NA (KeyError) shapes, and the follow-up ruff-format fix (head 8502428). CI is fully green on the current head. A re-review would be appreciated when there is time.

jbrockmendel · 2026-10-29T22:09:10Z


+    def test_setitem_iloc_rhs_series_non_default_index_with_na(self):
+        # GH#62473
+        # Assigning a nullable Int64 Series with an NA as the RHS of an iloc


can this comment be any more concise?

jbrockmendel · 2026-10-29T22:09:33Z

 - Bug in setitem (e.g. :meth:`Series.iloc`) silently storing wrong values when assigning a nullable or :class:`ArrowDtype` extension array into a NumPy-dtype column that could not hold it losslessly, such as a negative value into an unsigned-integer column (``-1`` became ``18446744073709551615``) or a value overflowing a narrower float dtype (:issue:`47776`)
 - Bug in setitem on nullable masked arrays (``Int64``, ``Float64``, ``boolean``, ...) silently coercing list-like values that the equivalent scalar setitem would reject — e.g. ``arr[[0]] = ['1']`` parsed the string into the numeric dtype, ``arr[[0]] = [True]`` on ``Int64``/``Float64`` coerced the boolean to ``1``, and an out-of-bounds value such as ``arr[[0]] = [1000]`` on ``Int8`` silently wrapped instead of raising (:issue:`45404`)
 - Fixed bug in :meth:`DataFrame.loc` where assigning an iterable to a single cell in an ``object`` dtype column incorrectly raised a ``ValueError`` (:issue:`26333`, :issue:`57962`)
+- Bug in :meth:`Series.iloc` setitem when the right-hand side is a nullable :class:`Series` (e.g. ``Int64``) with a non-default index, raising ``TypeError`` ("boolean value of NA is ambiguous") when the right-hand side contains a missing value or ``KeyError`` otherwise, because the coercion looked the assigned values up in the original by label instead of by position (:issue:`62473`)


needs to move to 3.2, also avoid discussing implementation details

jbrockmendel · 2026-10-30T17:08:05Z

+        # by label, which raised (or misfired) for a non-default index.
+        ser = pd.Series([1, 2], dtype="Int64")
+        result = ser.copy()
+        result.iloc[[0, 1]] = pd.Series([1.0, 2.0], index=["a", "b"])


i think this still passes even if the assignment totally failed? could we set values that dont happen to already match whats there?

…able Series RHS (GH#62473)

xirafa mentioned this pull request Sep 8, 2026

BUG: fix "boolean value of NA is ambiguous" in iloc setitem with nullable Series RHS autodag-dev/pandas#726

Open

github-actions Bot added the Awaiting Review PR in the maintainers' court; exempt from staleness label Sep 9, 2026

jbrockmendel reviewed Sep 14, 2026

View reviewed changes

jbrockmendel requested changes Sep 14, 2026

View reviewed changes

github-actions Bot removed the Awaiting Review PR in the maintainers' court; exempt from staleness label Sep 15, 2026

jbrockmendel requested changes Sep 28, 2026

View reviewed changes

github-actions Bot added the Stale label Oct 14, 2026

github-actions Bot removed the Stale label Oct 15, 2026

github-actions Bot added the Stale label Oct 29, 2026

github-actions Bot removed the Stale label Oct 30, 2026

jbrockmendel requested changes Oct 30, 2026

View reviewed changes

BUG: fix "boolean value of NA is ambiguous" in iloc setitem with null…

9ee9677

…able Series RHS (GH#62473)

IMGillusion force-pushed the fix-62473-iloc-setitem-nullable-na branch from 1ade738 to 9ee9677 Compare November 7, 2026 05:15

Conversation

IMGillusion commented Sep 8, 2026

Issue and Reproducible Example

Root Cause

Fix

Tests

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

IMGillusion commented Sep 14, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

IMGillusion commented Sep 29, 2026

Uh oh!

IMGillusion commented Sep 29, 2026

Uh oh!

github-actions Bot commented Oct 14, 2026

Uh oh!

IMGillusion commented Oct 14, 2026

Uh oh!

github-actions Bot commented Oct 29, 2026

Uh oh!

IMGillusion commented Oct 29, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants