search

david-cortes / contextualbandits

Python implementations of contextual bandits algorithms
http://contextual-bandits.readthedocs.io
BSD 2-Clause "Simplified" License
750 stars 148 forks source link

Getting topN arm features in Offpolicy method #58

Open shubh12et1062 opened 2 years ago

shubh12et1062 commented 2 years ago

Is it possible to get probabilities and topN arms in other methods like DoublyRobust and OffsetTree apart from Bootstrapped TS

david-cortes commented 2 years ago

It's not possible to obtain probabilities because the methods are not probabilistic. It should be theoretically possible to produce top-N rankings, but it's not currently implemented.

shubh12et1062 commented 2 years ago

Its possible through decision_function() to get scores of distributions across each arms. But I have observed that the mean score distribution across arms have very similar value.

shyun46 commented 1 year ago

I want to use linUCB (online policy) to my data. In my case, is decision_function()'s scores mean just the expected value of fitted model (not consider confidence bound)?

david-cortes commented 1 year ago

@shyun46 It includes the confidence bound.