Vectorized Predict for Non-contextual Policies

[bkleyn]

• Vectorize the pred_exp() function for non-contextual policies
• To enable multi-decision predict, allow non-contextual policies to accept contexts input (e.g., empty list) to indicate that multiple prediction should be generated
• Type-hinting and PEP8 clean up

[bkleyn]

one quick question, how does specifying none contexts enables vectorization?

Predictions will only be vectorized when contexts is not none. In terms of usage it looks like this.

mab = MAB([1, 2, 3], LearningPolicy.Random()) mab.fit(decisions=[1, 1, 2, 2], rewards=[0, 1, 1, 1]) predictions = mab.predict([[], [], []]) # generate 3 predictions

The number of predictions returned is equal to the length of the context input.