FIX possible UnboundLocalError in fetch_openml

[lesteve]

What does this implement/fix? Explain your changes.

The previous code looked like this:

try:
    X, y, frame, categories = _open_url_and_load_gzip_file(
        url, data_home, n_retries, delay, arff_params
    )
except Exception as exc:
    if parser == "pandas":
        from pandas.errors import ParserError

        if isinstance(exc, ParserError):
            # A parsing error could come from providing the wrong quotechar
            # to pandas. By default, we use a double quote. Thus, we retry
            # with a single quote before to raise the error.
            arff_params["read_csv_kwargs"] = {"quotechar": "'"}
            X, y, frame, categories = _open_url_and_load_gzip_file(
                url, data_home, n_retries, delay, arff_params
            )
        else:
            raise

return X, y, frame, categories

So if the first _open_url_and_load_gzip_file fails and parser != "pandas" we don't raise again in the except and we reach the return with X, y, frame, categories undefined causing an UnboundLocalError.

This was seen in a nightly build.

Any other comments?

I have not added a test for this, but let me know if you think this is necessary and I'll try to see how easy this is.

[thomasjpfan]

LGTM