- Update to 2.2.1
## Enhancements
* Added pyarrow pip extra so users can install pandas and pyarrow
with pip with pip install pandas[pyarrow] (#54466)
## Fixed regressions
* Fixed memory leak in `read_csv` (#57039)
* Fixed performance regression in `Series.combine_first` (#55845)
* Fixed regression causing overflow for near-minimum timestamps
(#57150)
* Fixed regression in `concat` changing long-standing behavior
that always sorted the non-concatenation axis when the axis was
a `DatetimeIndex` (#57006)
* Fixed regression in `merge_ordered` raising TypeError for
fill_method="ffill" and how="left" (#57010)
* Fixed regression in `pandas.testing.assert_series_equal`
defaulting to check_exact=True when checking the `Index`
(#57067)
* Fixed regression in `read_json` where an `Index` would be
returned instead of a `RangeIndex` (#57429)
* Fixed regression in `wide_to_long` raising an AttributeError
for string columns (#57066)
* Fixed regression in `.DataFrameGroupBy.idxmin`,
`.DataFrameGroupBy.idxmax`, `.SeriesGroupBy.idxmin`,
`.SeriesGroupBy.idxmax` ignoring the skipna argument (#57040)
* Fixed regression in `.DataFrameGroupBy.idxmin`,
`.DataFrameGroupBy.idxmax`, `.SeriesGroupBy.idxmin`,
`.SeriesGroupBy.idxmax` where values containing the minimum or
maximum value for the dtype could produce incorrect results
(#57040)
* Fixed regression in `CategoricalIndex.difference` raising
KeyError when other contains null values other than NaN
(#57318)
* Fixed regression in `DataFrame.groupby` raising ValueError when
grouping by a `Series` in some cases (#57276)
* Fixed regression in `DataFrame.loc` raising IndexError for
non-unique, masked dtype indexes where result has more than
10,000 rows (#57027)
* Fixed regression in `DataFrame.loc` which was unnecessarily
throwing "incompatible dtype warning" when expanding with
partial row indexer and multiple columns (see PDEP6) (#56503)
* Fixed regression in `DataFrame.map` with na_action="ignore" not
being respected for NumPy nullable and `ArrowDtypes` (#57316)
* Fixed regression in `DataFrame.merge` raising ValueError for
certain types of 3rd-party extension arrays (#57316)
* Fixed regression in `DataFrame.query` with all NaT column with
object dtype (#57068)
* Fixed regression in `DataFrame.shift` raising AssertionError
for axis=1 and empty `DataFrame` (#57301)
* Fixed regression in `DataFrame.sort_index` not producing a
stable sort for a index with duplicates (#57151)
* Fixed regression in `DataFrame.to_dict` with orient='list' and
datetime or timedelta types returning integers (#54824)
* Fixed regression in `DataFrame.to_json` converting nullable
integers to floats (#57224)
* Fixed regression in `DataFrame.to_sql` when method="multi" is
passed and the dialect type is not Oracle (#57310)
* Fixed regression in `DataFrame.transpose` with nullable
extension dtypes not having F-contiguous data potentially
causing exceptions when used (#57315)
* Fixed regression in `DataFrame.update` emitting incorrect
warnings about downcasting (#57124)
* Fixed regression in `DataFrameGroupBy.idxmin`,
`DataFrameGroupBy.idxmax`, `SeriesGroupBy.idxmin`,
`SeriesGroupBy.idxmax` ignoring the skipna argument (#57040)
* Fixed regression in `DataFrameGroupBy.idxmin`,
`DataFrameGroupBy.idxmax`, `SeriesGroupBy.idxmin`,
`SeriesGroupBy.idxmax` where values containing the minimum or
maximum value for the dtype could produce incorrect results
(#57040)
* Fixed regression in `ExtensionArray.to_numpy` raising for
non-numeric masked dtypes (#56991)
* Fixed regression in `Index.join` raising TypeError when joining
an empty index to a non-empty index containing mixed dtype
values (#57048)
* Fixed regression in `Series.astype` introducing decimals when
converting from integer with missing values to string dtype
(#57418)
* Fixed regression in `Series.pct_change` raising a ValueError
for an empty `Series` (#57056)
* Fixed regression in `Series.to_numpy` when dtype is given as
float and the data contains NaNs (#57121)
* Fixed regression in addition or subtraction of `DateOffset`
objects with millisecond components to datetime64 `Index`,
`Series`, or `DataFrame` (#57529)
## Bug fixes
* Fixed bug in `pandas.api.interchange.from_dataframe` which was
raising for Nullable integers (#55069)
* Fixed bug in `pandas.api.interchange.from_dataframe` which was
raising for empty inputs (#56700)
* Fixed bug in `pandas.api.interchange.from_dataframe` which
wasn't converting columns names to strings (#55069)
* Fixed bug in `DataFrame.__getitem__` for empty `DataFrame` with
Copy-on-Write enabled (#57130)
* Fixed bug in `PeriodIndex.asfreq` which was silently converting
frequencies which are not supported as period frequencies
instead of raising an error (#56945)
## Note
* The DeprecationWarning that was raised when pandas was imported
without PyArrow being installed has been removed. This decision
was made because the warning was too noisy for too many users
and a lot of feedback was collected about the decision to make
PyArrow a required dependency. Pandas is currently considering
the decision whether or not PyArrow should be added as a hard
dependency in 3.0. Interested users can follow the discussion
here.
* Added the argument skipna to `DataFrameGroupBy.first`,
`DataFrameGroupBy.last`, `SeriesGroupBy.first`, and
`SeriesGroupBy.last`; achieving skipna=False used to be
available via `DataFrameGroupBy.nth`, but the behavior was
changed in pandas 2.0.0 (#57019)
* Added the argument skipna to `Resampler.first`,
`Resampler.last` (#57019)
- Release notes for 2.2.0
* For full changelog see
https://github.com/pandas-dev/pandas/blob/main/doc/source/whatsnew/v2.2.0.rst
## Enhancements
* ADBC Driver support in to_sql and read_sql
* Create a pandas Series based on one or more conditions
* to_numpy for NumPy nullable and Arrow types converts to
suitable NumPy dtype
* Series.struct accessor for PyArrow structured data
* Series.list accessor for PyArrow list data
* Calamine engine for `read_excel`
## Notable bug fixes
* `merge` and `DataFrame.join` now consistently follow documented
sort behavior
* `merge` and `DataFrame.join` no longer reorder levels when
levels differ
* Increased minimum versions for dependencies
## Deprecations
* Chained assignment
* Deprecate aliases M, Q, Y, etc. in favour of ME, QE, YE, etc.
for offsets
* Deprecated automatic downcasting
- Simplify flavor test setup: obs can evaluate %{shrink:} now