Code for reproducing results in the paper "Thresholded Lasso Bandit".
In this paper, we revisit the regret minimization problem in sparse stochastic contextual linear bandits, where feature vectors may be of large dimension
This code is written in Python 3. To install the required dependencies, execute the following command:
$ pip install -r requirements.txt
Build the container:
$ docker build -t thresholded-lasso-bandit .
After build finished, run the container:
$ docker run -it thresholded-lasso-bandit
In order to investigate the performance of TH Lasso bandit on features drawn from a Gaussian distribution, execute the following command:
$ python run_gaussian_experiment.py
In this experiment, the following options can be specified:
--K
: Number of arms. The default value is2
.--T
: Number of rounds to be played. The default value is1000
.--d
: Dimension of feature vectors. The default value is1000
.--s0
: Sparsity index. The default value is20
.--x_max
: Maximum l2-norm of feature vectors. The default value is10
.--rho_sq
: Correlation level between feature vectors of arms. The default value is0.7
.--num_trial
: Number of trials to run experiments. The default value is20
.
To evaluate TH Lasso bandit via an experiment with a feature distribution other than the Gaussian distribution (uniform, elliptical, hard instance), execute the following command:
$ python run_uniform_experiment.py
$ python run_elliptical_experiment.py
$ python run_hard_instance_experiment.py
If you use our code in your work, please cite our paper:
@InProceedings{ariu2022thlassobandit,
title = {Thresholded Lasso Bandit},
author = {Ariu, Kaito and Abe, Kenshi and Proutiere, Alexandre},
booktitle = {Proceedings of the 39th International Conference on Machine Learning},
pages = {878--928},
year = {2022},
volume = {162}
}