MaxHistoryTrackerFeaturizer featurizer creates a feature vector (num_unique_turns, max_history, num_input_features).
Now I understand where the max_history parameter and the num_input_features come from.
However, it is still a mystery for num_input_features.
Any hints?
I assume in the question, you meant what is num_unique_turns. It is a number of training examples, or in the context of rasa_core it is a number of actions to be predicted (we consider an action predicted by rasa_core as a turn of the dialogue). We call it unique because by default the featurizer performs deduplication based on the corresponding feature vectors