This repo contains the implementation of a nonstationary bandit learning environment and a state-of-the-art non-stationary contextual bandit algorithm dLinUCB, which is published in SIGIR2018 [1]. (More related non-stationary algorithms will be added, stay tuned!).

##Usage Run the simulator: python NonStationaryEnv_Simulation.py --alg XXX where parameter alg represents the name of algorithm.

Run with different parameters: python NonStationaryEnv_Simulation.py --alg XXX --tau XX --delta_1 XX --delta_2 XX where --tau, --delta_1, --delta_2 are the input parameters of dLinUCB

##Algorithms' details LinUCB: A state-of-the-art contextual bandit algorithm. It select arms based on an upper confidence bound of the estimated reward with given context vectors. LinUCB assume that users/bandits' parameters are independent with each other. And LinUCB only works with the observed features and does not consider hidden features.

dLinUCB: A state-of-the-art non-stationary contextual bandit algorithm that detects possible changes of environment based on its reward estimation confidence and updates its arm selection strategy respectively.

##Result The following is a sample result by running 'python NonStationaryEnv_Simulation.py'

##Reference [1]: Qingyun Wu, Naveen Iyer, and Hongning Wang. Learning Contextual Bandits in a Non-stationary Environment. The 41th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR'2018), https://doi.org/10.1145/3209978.3210051