Support for ordinal classification or regression-based models in the pipeline

[zachschillaci27]

The TextCategorizer component is ill suited in cases where there is an ordering to the classes or a continuous variable to be predicted. Therefore, support could be added for a TextOrdinalCategorizer or TextRegressor pipeline to predict
an ordinal classification or a continuous response variable, respectively. Such a pipeline could be used for predicting classic ordinal classification problems such as movie star ratings or regressor problems such as sentiment analysis.

[koaning]

Part of me wonders if you'd really need a separate categorizer for this. In the case of a 5-star review rating, you can normalize the values of 1-5 to a single classifier that needs to predict a value between [0, 1]. It's a trick that I use here. I've not tried this approach using spaCy, but conceptually it might be simpler to first re-use the existing tools.