Statistical Efficiency of Thompson Sampling for Combinatorial Semi-Bandits

Authors: Pierre Perrault, Etienne Boursier, Michal Valko, Vianney Perchet

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Experiments, We compare our CTS policies to CUCB and CUCB-KL, for the shortest path problem on the road chesapeake network [Rossi and Ahmed, 2015]., Our results are shown in Figure 2, where we observe that CLIP CTS-GAUSSIAN (resp. ESCB) is slightly better for c small (resp. large), thus reaching the best of both worlds., Table 2: Computation time per round (ms)
Researcher Affiliation Collaboration Pierre Perrault Inria Lille ENS Paris-Saclay pierre.perrault@inria.fr Etienne Boursier ENS Paris-Saclay etienne.boursier1@gmail.com Vianney Perchet ENSAE Criteo AI Lab vianney.perchet@normalesup.org Michal Valko Deep Mind Paris Inria Lille valkom@deepmind.com
Pseudocode Yes Algorithm 1 CTS-BETA, Algorithm 2 CTS-GAUSSIAN
Open Source Code No The paper does not contain any explicit statements or links indicating that the source code for the described methodologies is publicly available.
Open Datasets Yes shortest path problem on the road chesapeake network [Rossi and Ahmed, 2015]
Dataset Splits No The paper describes an online learning setting and does not specify traditional train/validation/test dataset splits. It mentions averaged over 50 simulations but not data partitioning.
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments, such as GPU/CPU models or memory specifications.
Software Dependencies No The paper does not specify any software dependencies with version numbers, such as programming languages, libraries, or specialized solvers.
Experiment Setup Yes Before describing the experiments carried out, notice that in the CTS-GAUSSIAN policies, β > 1 is an artefact of the analysis and can in practice be taken equal to 1. This is what we did in our experiments.