[ENH] sklearn facing coercion utility for pd.DataFrame, to str …

…columns (#5550)

This PR adds a `sklearn` facing coercion utility for `pd.DataFrame`, to
`str` columns, and tests for it.

As this is a second small module with `sklearn` themed functionality,
the current module and its tests are moved one folder down together with
the new content.

sktime/utils/sklearn/__init__.py

@@ -0,0 +1,21 @@
+"""Sklearn related utility functionality."""
+
+from sktime.utils.sklearn._adapt_df import prep_skl_df
+from sktime.utils.sklearn._scitype import (
+    is_sklearn_classifier,
+    is_sklearn_clusterer,
+    is_sklearn_estimator,
+    is_sklearn_regressor,
+    is_sklearn_transformer,
+    sklearn_scitype,
+)
+
+__all__ = [
+    "prep_skl_df",
+    "is_sklearn_estimator",
+    "is_sklearn_transformer",
+    "is_sklearn_classifier",
+    "is_sklearn_regressor",
+    "is_sklearn_clusterer",
+    "sklearn_scitype",
+]

sktime/utils/sklearn/_adapt_df.py

@@ -0,0 +1,31 @@
+"""Utility functions for adapting to sklearn."""
+# copyright: sktime developers, BSD-3-Clause License (see LICENSE file)
+
+import numpy as np
+
+
+def prep_skl_df(df, copy_df=False):
+    """Make df compatible with sklearn input expectations.
+
+    Changes:
+    turns column index into a list of strings
+
+    Parameters
+    ----------
+    df : pd.DataFrame
+        list of indices to sample from
+    copy_df : bool, default=False
+        whether to mutate df or return a copy
+        if False, index of df is mutated
+        if True, original df is not mutated. If index is not a list of strings,
+        a copy is made and the copy is mutated. Otherwise, the original df is returned.
+    """
+    cols = df.columns
+    str_cols = cols.astype(str)
+
+    if not np.all(str_cols == cols):
+        if copy_df:
+            df = df.copy()
+        df.columns = str_cols
+
+    return df

sktime/utils/sklearn/tests/__init__.py

@@ -0,0 +1 @@
+"""Tests for sklearn related utility functionality."""

sktime/utils/sklearn/tests/test_sklearn_df_adapt.py

@@ -0,0 +1,33 @@
+"""Tests for sklearn dataframe coercion."""
+
+__author__ = ["fkiraly"]
+
+import numpy as np
+import pandas as pd
+import pytest
+
+from sktime.utils.sklearn._adapt_df import prep_skl_df
+
+
+@pytest.mark.parametrize("copy_df", [True, False])
+def test_prep_skl_df_coercion(copy_df):
+    """Test that prep_skl_df behaves correctly on the coercion case."""
+    mixed_example = pd.DataFrame({0: [1, 2, 3], "b": [1, 2, 3]})
+
+    res = prep_skl_df(mixed_example, copy_df=copy_df)
+
+    assert np.all(res.columns == ["0", "b"])
+
+    if not copy_df:
+        assert res is mixed_example
+
+
+@pytest.mark.parametrize("copy_df", [True, False])
+def test_prep_skl_df_non_coercion(copy_df):
+    """Test that prep_skl_df behaves correctly on the non-coercion case."""
+    mixed_example = pd.DataFrame({"a": [1, 2, 3], "b": [1, 2, 3]})
+
+    res = prep_skl_df(mixed_example, copy_df=copy_df)
+
+    assert np.all(res.columns == ["a", "b"])
+    assert res is mixed_example