streaming-regret-minimization-MABs

Codes for the experiments in paper 'Tight Regret Bounds for Single-pass Streaming Multi-armed Bandits.'

Run test.py or test.ipynb to get test the performances of the implemented algorithms.

Change the variable num_arms to adjust the number of arms ($K$); change trial_mult_factor ($\alpha$) and trial_exp_factor ($\beta$) to change the relationship between $K$ and $T$, i.e. $T=\alpha \cdot K^{\beta}$.

Change the arm_setting variable between clear_cut_setting and mix_in_setting to control the setting of the stream, i.e. one arm with much higher mean reward vs. arms with similar rewards.