We're excited to announce the release of pandas 3.0.0. This major long-awaited release brings significant improvements to pandas, but also features some potentially breaking changes.

Highlights of pandas 3.0

pandas 3.0 introduces several major enhancements:

• Dedicated string data type by default: string columns are now inferred as the new str dtype instead of object, providing better performance and type safety
• Consistent copy/view behaviour with Copy-on-Write (CoW) (a.k.a. getting rid of the SettingWithCopyWarning): more predictable and consistent behavior for all operations, with improved performance through avoiding unnecessary copies
• New default resolution for datetime-like data: no longer defaulting to nanoseconds, but generally microseconds (or the resolution of the input), when constructing datetime or timedelta data (avoiding out-of-bounds errors for dates with a year before 1678 or after 2262)
• New pd.col syntax: initial support for pd.col() as a simplified syntax for creating callables in DataFrame.assign

Further, pandas 3.0 includes a lot of other improvements and bug fixes. You can find the complete list of changes in the release notes.

Upgrading to pandas 3.0

The pandas 3.0 release removed functionality that was deprecated in previous releases (see here for an overview). It is recommended to first upgrade to pandas 2.3 and to ensure your code is working without warnings, before upgrading to pandas 3.0.

Further, as a major release, pandas 3.0 includes some breaking changes that may require updates to your code. The two most significant changes are the new string dtype and the copy/view behaviour changes, detailed below. An overview of all potentially breaking changes can be found in the Backwards incompatible API changes section of the release notes.

1. Dedicated string data type by default

Starting with pandas 3.0, string columns are automatically inferred as str dtype instead of the numpy object (which can store any Python object).

Example:

# Old behavior (pandas < 3.0)
>>> ser = pd.Series(["a", "b"])
>>> ser
0    a
1    b
dtype: object  # <-- numpy object dtype

# New behavior (pandas 3.0)
>>> ser = pd.Series(["a", "b"])
>>> ser.dtype
>>> ser
0    a
1    b
dtype: str  # <-- new string dtype

This change improves performance and type safety, but may require code updates, especially for library code that currently looks for "object" dtype when expecting string data.

For more details, see the migration guide for the new string data type.

This new data type will use the pyarrow library under the hood, if installed, to provide the performance improvements. Therefore we strongly recommend to install pyarrow alongside pandas (but pyarrow is not a required dependency installed by default).

2. Consistent copy/view behaviour with Copy-on-Write (CoW)

Copy-on-Write is now the default and only mode in pandas 3.0. This makes behavior more consistent and predictable, and avoids a lot of defensive copying (improving performance), but requires updates to certain coding patterns.

The most impactfull change is that chained assignment will no longer work. As a result, the SettingWithCopyWarning is also removed (since there is no longer ambiguity whether it would work or not), and defensive .copy() calls to silence the warning are no longer needed.

Example:

# Old behavior (pandas < 3.0) - chained assignment
df["foo"][df["bar"] > 5] =   # This might modify df (unpredictable)

# New behavior (pandas 3.0) - must do the modification in one step (e.g. with .loc)
df.loc[df["bar"] > 5, "foo"] = 100

In general, any result of an indexing operation or method now always behaves as if it were a copy, so modifications of the result won't affect the original DataFrame.

For more details, see the Copy-on-Write migration guide.

Obtaining pandas 3.0

You can install the latest pandas 3.0 release from PyPI:

python -m pip install --upgrade pandas==3.0.*

Or from conda-forge using conda/mamba:

conda install -c conda-forge pandas=3.0

Running into an issue or regression?

Please report any problem you encounter with the release on the pandas issue tracker.

Thanks to all the contributors who made this release possible!