This repository aims to reproduce and expend the "Batched Bandits Problems" and extend it by adding the following experiments:
- Ploting the "optimal grid" as suggested in section 4.2
- PredTau algorithm which estimate delta and predict tau(\Delta)
- Improved-UCB - I added this experiment for fair comparison
- Improved-UCB with go-to-broke policy
- Batched Arm Elimination algorithm based on the paper "Regret Bounds for Batched Bandits"
This is not the official code of the paper
The code was implementedin python3 and uses scipy and numpy. Please install using pip:
pip install numpy
pip install scipy
python main_paralle --alg improved_ucb_gtb --d poisson
python ploter --sub_plot