Learning to Bid without Knowing your Value
Code for reproducing experimental results from our paper: https://arxiv.org/pdf/1711.01333.pdf.
- Python 2.7.13
- conda 4.4.10
- RStudio (if you want to run the plotters)
Folders for different simulations with oblivious adveraries (stochastic opponents), adaptive adversaries using EXP3 and adaptive adversaries using WINEXP. In order to run each simulation:
This will create 2 .txt files: winexp_regrets.txt and exp3_regrets.txt. These files hold the regrets for each repetition and each timestep and they are of size num_repetitions x T. Further, it will create a .png file, with the plot of regrets.
For the folders that create the graphs for different discretizations there is a separate .txt file for each algorithm (winexp and exp3) and each discretization (0.001, 0.01, 0.1).
We also include .R files that extract the graphs along with the final .png that appear in our paper.