fix(skore): convert DataFrame column names to string type

[waridrox]

Closes #2029
CC: @thomass-dev

When debugging, in the data_accessor.py file:

skore/skore/src/skore/_sklearn/_estimator/data_accessor.py

Lines 49 to 67 in eb0d6e9

	if X is None:
	raise ValueError(err_msg.format(f"X_{dataset}", data_source))
	elif not sbd.is_dataframe(X):
	X = pd.DataFrame(X, columns=[f"Feature {i}" for i in range(X.shape[1])])
	if with_y:
	if y is None:
	raise ValueError(err_msg.format(f"y_{dataset}", data_source))
	if isinstance(y, pd.Series) and y.name is not None:
	y = y.to_frame()
	elif not sbd.is_dataframe(y):
	if y.ndim == 1:
	columns = ["Target"]
	else:
	columns = [f"Target {i}" for i in range(y.shape[1])]
	y = pd.DataFrame(y, columns=columns)
	return X, y

This method only converts numpy arrays to DataFrames with string column names, but doesn't handle the case if DataFrames already exist but have integer type column names.

This then further passes down to the skrub functions (which expects strings), and the final error of TypeError: cannot use a string pattern on a bytes-like object occurs because suggested_name is an integer from the RangeIndex columns.

def _get_new_name(suggested_name, forbidden_names):
    """Get a new name for a column."""
    # .......    
    tags = re.findall(tag_pattern, suggested_name)  # <==== 
    # .......

tag_pattern is a string regex pattern and re.findall(tag_pattern, suggested_name) method expects both arguments to be strings or both to be bytes-like objects.

I tried to overcome this by ensuring that the DataFrame object has string column names if it already exists to avoid issues with skrub later when passing down from data_accessor.py function.

Alternatively, should I simply just raise exception errors if the compute fails in subsequent steps instead of this?

[thomass-dev]

Thanks for your investigation and your contribution. Please add a dedicated test.

[waridrox]

This test verifies that when X/y are DataFrames with integer RangeIndex columns on a multiclass dataset, report.data.analyze() runs and the resulting columns are strings.

[waridrox]

This test verifies that mixed column name types in X (ints and strings) and integer column in y are normalized so analyze(data_source="all") succeeds and output columns are strings.