Hi, When using using date_features, such as: 'quarter', 'month', 'week', 'dayofweek', 'day', 'hour', day_occurrence, in MLForecast, does the model read the features as ordinal features, or does it account for the fact that hour 0 for example and hour 23 are close ?

It just computes the feature values. You can create a scikit-learn pipeline that does some processing before passing them to the model