Structural Causal Bandits: Where to Intervene?

Authors: Sanghack Lee, Elias Bareinboim

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we present empirical results demonstrating that the selection of arms based on POMISs makes standard MAB solvers converge faster to an optimal arm.
Researcher Affiliation Academia Sanghack Lee Department of Computer Science Purdue University lee2995@purdue.edu Elias Bareinboim Department of Computer Science Purdue University eb@purdue.edu
Pseudocode Yes Algorithm 1 Algorithm enumerating all POMISs with JG, Y K
Open Source Code Yes All the code is available at https://github.com/sanghack81/SCMMAB-NIPS2018
Open Datasets No The paper uses synthetic SCM-MAB instances that are generated and parameterized according to specifications in Appendix D, rather than referring to a named, publicly available dataset with concrete access information or citation.
Dataset Splits No The paper describes a multi-armed bandit simulation setup and does not refer to traditional train/validation/test dataset splits, which are not applicable in this context.
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, memory, or processing units used for running the simulations.
Software Dependencies No The paper mentions using 'kl-UCB' and 'Thompson sampling' algorithms, but it does not specify any software dependencies with version numbers (e.g., Python version, library versions).
Experiment Setup Yes We set the horizon large enough so as to observe near convergence, and repeat each simulation 300 times. We set the horizon T = 1000 for Task 1 and Task 2, and T = 10000 for Task 3. We repeat each simulation 300 times.