Stochastic bandits with groups of similar arms.

Authors: Fabien Pesquerel, Hassan SABER, Odalric-Ambrym Maillard

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Last, we illustrate the performance of the considered strategy on numerical experiments involving a large number of arms. In this section, we support our theoretical analysis by conducting three sets of experiments. The Python code used to perform those experiments is available on Github. In this section, all the experiments are conducted using gaussian distributions whose means are between 0 and 1 and of unit standard deviation. Those graphs are representative of all the experiments that we conducted and more plots and experiments may be found in the appendix D.
Researcher Affiliation Academia Univ. Lille, CNRS, Inria, Centrale Lille, UMR 9198-CRISt AL, F-59000 Lille, France
Pseudocode Yes The IMED-EC algorithm is presented in Algorithm 1. Algorithm 1 IMED-EC (IMED for Equivalent Classes)
Open Source Code Yes The Python code used to perform those experiments is available on Github2. 2https://github.com/fabienpesquerel/stochastic-bandits-with-groups-of-similar-arms-neurips-2021
Open Datasets No The paper describes generating data for its experiments using
Dataset Splits No The paper describes experimental setups with varying numbers of classes and distributions, but it does not specify explicit training, validation, or test dataset splits. The problem is framed as a sequential decision-making problem (multi-armed bandit), where traditional dataset splits are not typically used.
Hardware Specification No This type of experiment does not take more than roughly 10 to 15 minutes on a notebook run in Google Colab depending on the number of arms, the horizon and the number of runs. This supports the numerical efficiency of the relaxation made in IMED-EC.
Software Dependencies No The paper mentions "The Python code used to perform those experiments is available on Github" but does not specify the version of Python or any other software dependencies like libraries or frameworks with their versions.
Experiment Setup Yes In this section, all the experiments are conducted using gaussian distributions whose means are between 0 and 1 and of unit standard deviation. Figure 1: 3 classes, 3 distributions per class set of means = {0.3, 0.5, 0.9}. Figure 2: 7 classes, 8 distributions per class set of means = {0.1, 0.3, 0.4, 0.5, 0.6, 0.75, 0.9}. The experiment Figure 3 is performed on a bandit problem with 7 classes and an uneven number of distributions per class. The smallest class has 4 elements and the largest 23.