search

david-cortes / contextualbandits

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

Single Model #22

Closed htcml closed 4 years ago

htcml commented 4 years ago

Are there any implemented algorithms in this library which uses a single unified model for all arms?

david-cortes commented 4 years ago

No, there isn't any yet. Also the library doesn't use arm features.

htcml commented 4 years ago

@david-cortes Since you are an expert in bandit algorithms, can you point out some papers about unified models? Is it worthwhile to pursue the unified model approach? The only paper I came across is this one: https://arxiv.org/pdf/1810.09558.pdf

david-cortes commented 4 years ago

No idea. From my own experiments, when trying to do a single model with arm features and no similarity info, I haven't got any uplift over a naive policy which would always use argmax with no exploration strategy.