Known bug in scikit-learn's f1_score metric

[agmoore4]

Related to #29734 (scikit-learn/scikit-learn#29734).

Scikit-learn's documentation states that the pos_label parameter is ignored in multiclass problems, but it isn't actually ignored. Since it defaults to pos_label=1, which isn't a label found in the dataset, the scoring fails every time, giving np.nan.

Setting pos_label=None explicitly would resolve the problem and have f1_score work as intended.

[agmoore4]

This fix would be relevant in tutorials/text_class_example.py, which is used in the text module tutorial