-
Notifications
You must be signed in to change notification settings - Fork 2
Open
Labels
Priority/1-MediumTo do after P0To do after P0Size/HoursVery shortVery shortStatus/ReadyForDevThe issue is ready to be developed or to be investigated deeplyThe issue is ready to be developed or to be investigated deeply
Milestone
Description
Pandas 3.0.0 has been released on January 21st, 2026. Hence, all environments that use Python >= 3.11 pull Pandas 3.0.0 as a dependency of khiops-python.
However, Pandas 3.0.0 contains breaking changes with respect to previous versions, including Pandas 2.3.3. Most notably, the StringDType is used by default, rather than object. As the StringDType is a Pandas data type, as opposed to object which is a NumPy data type, all code assuming a by-default Pandas / NumPy compatibility regarding the Pandas column types breaks now. For example, code like np.issubdtype(my_pandas_dtype, numpy_dtype) breaks if my_pandas_dtype is a StringDType, with TypeError: Cannot interpret '<StringDtype(storage='python', na_value=nan)>' as a data type.
Metadata
Metadata
Assignees
Labels
Priority/1-MediumTo do after P0To do after P0Size/HoursVery shortVery shortStatus/ReadyForDevThe issue is ready to be developed or to be investigated deeplyThe issue is ready to be developed or to be investigated deeply