BUG: Fix numeric_only ignored with list of functions in DataFrame.agg and GroupBy.agg (#49352)

[ali-aqib]

Fixes #49352

This PR implements the filtering approach suggested by @rhshadrach in issue #49352, making numeric_only=True work correctly when passing a list of aggregation functions to DataFrame.agg() and GroupBy.agg().

---

Problem

When calling .agg() with a list of functions and numeric_only=True, the parameter was being ignored, causing errors or incorrect behavior:

import pandas as pd

df = pd.DataFrame({
    'A': [1, 2, 3, 4, 5],
    'B': [10.5, 20.5, 30.5, 40.5, 50.5],
    'C': ['a', 'b', 'c', 'd', 'e']
})

# Before: Would fail with TypeError
print(df.agg(['min', 'max', 'mean'], numeric_only=True))
# TypeError: Series.min does not allow numeric_only=True with non-numeric dtypes.
# Error occurs because numeric_only=True is passed to string column 'C'

# GroupBy had the same issue
df2 = pd.DataFrame({
    'key': ['A', 'B', 'A', 'B'],
    'num': [1, 2, 3, 4],
    'text': ['a', 'b', 'c', 'd']
})

print(df2.groupby('key').agg(['sum', 'mean'], numeric_only=True))
# TypeError: Cannot use numeric_only=True with SeriesGroupBy.sum and non-numeric dtypes.
# numeric_only parameter was ignored, causing operations on string columns

---

Root Cause

As identified by @rhshadrach in this comment:

"We should be applying numeric_only=True in df.agg prior to splitting the DataFrame into Series. We then also need to not pass numeric_only=True to the functions when they are called on the Series."

The issue occurred because:

1. numeric_only was not being intercepted before splitting the DataFrame into Series
2. When passed to Series methods, it caused TypeError (Series aggregation methods don't accept numeric_only)

---

Solution

This PR implements Option 1 from the discussion:

"Intercept numeric_only arguments and apply them to the DataFrame prior to splitting the data up into Series"

Implementation Details:

1. In NDFrameApply.agg_or_apply_list_like() (for DataFrame.agg):

   • Check if numeric_only=True is passed
   • Filter DataFrame to numeric columns using select_dtypes(include="number")
   • Remove numeric_only from kwargs before passing to Series methods
   • Process only numeric columns

2. In GroupByApply.agg_list_like() (for GroupBy.agg):

   • Apply the same filtering logic
   • Ensures consistency between DataFrame and GroupBy behavior

Why This Approach?

• ✅ Simple: Minimal code changes, leverages existing select_dtypes()
• ✅ Consistent: Follows pandas' existing filtering philosophy
• ✅ No Breaking Changes: Uses existing numeric_only parameter
• ✅ Avoids kwargs issues: Doesn't pass numeric_only through to user functions (learned from API: add numeric_only support to groupby agg #58132)

---

After This PR

import pandas as pd

df = pd.DataFrame({
    'A': [1, 2, 3, 4, 5],
    'B': [10.5, 20.5, 30.5, 40.5, 50.5],
    'C': ['a', 'b', 'c', 'd', 'e']
})

# Works correctly - excludes column 'C'
result = df.agg(['min', 'max', 'mean'], numeric_only=True)
print(result)
#        A     B
# min   1.0  10.5
# max   5.0  50.5
# mean  3.0  30.5

# GroupBy also works
df2 = pd.DataFrame({
    'key': ['A', 'B', 'A', 'B'],
    'num': [1, 2, 3, 4],
    'text': ['a', 'b', 'c', 'd']
})

result = df2.groupby('key').agg(['sum', 'mean'], numeric_only=True)
# Only 'num' column is aggregated, 'text' is excluded ✅
print(result)
#    num      
#    sum mean
# key        
# A    4  2.0
# B    6  3.0

---

Related Issues and PRs

• Primary Issue: BUG: inconsistent DataFrame.agg behavoir when passing as kwargs numeric_only=True #49352 (this fix)
• Related Discussion: API: Consistently support numeric_only in groupby ops #56946 (broader numeric_only support in groupby)
• Previous Attempt: API: add numeric_only support to groupby agg #58132 (closed due to kwargs-passing approach issues)
• Related Bug: BUG: agg ignores args/kwargs when giving a list-like argument #50624 (kwargs not passed through for list functions - separate issue)

This PR avoids the problems encountered in #58132 by:

• Not passing numeric_only via **kwargs to user functions
• Using simple filtering instead of complex caching mechanisms
• Targeting the specific list-of-functions code path

---

Testing

New test files added:

• pandas/tests/apply/test_frame_apply_numeric_only.py (28 tests)
• pandas/tests/groupby/aggregate/test_aggregate_numeric_only.py (21 tests)

Test coverage includes:

• Mixed dtypes (numeric + non-numeric columns)
• All numeric columns
• No numeric columns (edge case)
• Various function types (string, numpy functions, mixed)
• Different numbers of functions
• Single vs multiple grouping columns
• Parameter variations (numeric_only=True/False)
• Column order preservation
• Edge cases (NaNs, datetime columns, single row, large DataFrame)
• as_index parameter for GroupBy

All tests pass locally:

pytest pandas/tests/apply/test_frame_apply_numeric_only.py -v  # 28 passed
pytest pandas/tests/groupby/aggregate/test_aggregate_numeric_only.py -v  # 21 passed

---

Backward Compatibility

✅ Fully backward compatible

• Uses existing numeric_only parameter
• No API changes
• Existing behavior preserved for non-list functions
• Only affects previously broken/inconsistent list-of-functions case

---

cc

@rhshadrach - This implements the interception approach you suggested in #49352. Would appreciate your review!

---

Checklist

• closes BUG: inconsistent DataFrame.agg behavoir when passing as kwargs numeric_only=True #49352
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed
• Added type annotations to new arguments/methods/functions (N/A - no new functions added)
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature