has_dtype

skrub.selectors.has_dtype(*dtypes)

Select columns whose dtype is equal to one of the provided dtypes.

This is an advanced selector for edge cases where you need to match specific dtypes not covered by other selectors. Use this when working with specialized or custom dtypes (e.g., pandas ListDtype, polars Object).

For standard types, prefer the simpler selectors like numeric(), string(), categorical(), or boolean().

This selector takes a hands-off approach: skrub does not normalize or infer dtypes across dataframe libraries. A column is selected if sbd.dtype(column) == dtype for at least one of the provided dtypes.

Parameters:

*dtypesdtype objects

One or more dtype objects to match.

See also

numeric

Select numeric columns.

string

Select string columns.

categorical

Select categorical columns.

object

Select columns with “object” dtype (library specific).

Notes

Some dataframe libraries may accept shorthand values that compare equal to a dtype object (e.g., ‘int64’), but this is backend-specific. For robustness, pass dtype objects obtained from your dataframe library.

Examples

>>> from skrub import selectors as s
>>> import pandas as pd
>>> df = pd.DataFrame(
...     {
...         "items": [["A4", "A3"], ["A5"]],
...         "count": [2, 1],
...     }
... )

Get dtype from an existing column and use it for selection:

>>> s.select(df, s.has_dtype(df["items"].dtype))
       items
0  [A4, A3]
1      [A5]

Match multiple dtypes at once:

>>> items_dtype = df["items"].dtype
>>> count_dtype = df["count"].dtype
>>> s.select(df, s.has_dtype(items_dtype, count_dtype))
       items  count
0  [A4, A3]      2
1      [A5]      1

This also works with complex dtypes such as GeometryDtype in GeoPandas:

>>> import geopandas as gpd
>>> from shapely.geometry import Point
>>> gdf = gpd.GeoDataFrame(
...     {"city": ["Paris", "Berlin"], "value": [1, 2]},
...     geometry=[Point(2.35, 48.86), Point(13.40, 52.52)],
...     crs="EPSG:4326",
... )
>>> gdf
    city  value            geometry
0   Paris      1  POINT (2.35 48.86)
1  Berlin      2  POINT (13.4 52.52)
>>> s.select(gdf, s.has_dtype(gdf.geometry.dtype))
            geometry
0  POINT (2.35 48.86)
1  POINT (13.4 52.52)