Structural Causal Bandits with Non-Manipulable Variables

Authors: Sanghack Lee, Elias Bareinboim4164-4172

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We corroborate our findings with simulations, which shows that MAB solvers enhanced with causal knowledge and leveraging the newly discovered dependence structure among arms consistently outperform causal-insensitive solvers. (...) We simulated three SCM-MAB instances, i.e., the front-door setting (Fig. 1) and two models following Fig. 2a and Fig. 5d. The time horizon T is set to 10,000, which is enough to observe the performance difference; simulations are repeated 1,000 times, and the number of bootstraps B is set to 500.
Researcher Affiliation Academia Sanghack Lee, Elias Bareinboim Department of Computer Science Purdue University West Lafayette, IN 47907 {lee2995,eb}@purdue.edu
Pseudocode Yes Algorithm 1 z2ID (...) Algorithm 2 b MVWA (...) Algorithm 3 Bernoulli TS and KL-UCB with z2ID
Open Source Code No The paper does not provide any explicit statement about open-sourcing its code or a link to a code repository.
Open Datasets No The paper states "We simulated three SCM-MAB instances" for its experiments. It does not mention using any publicly available datasets or provide access information for any dataset used.
Dataset Splits No The paper conducts simulations of a bandit problem and discusses concepts like 'time horizon' and 'bootstrap samples' for variance estimation. However, it does not specify explicit training, validation, or test dataset splits as would be typical for experiments on fixed datasets.
Hardware Specification No The paper mentions that the experiments were 'simulated' but does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used to run these simulations.
Software Dependencies No The paper mentions using MAB algorithms like TS and KL-UCB, but it does not specify any software names with version numbers (e.g., Python, PyTorch, specific libraries or solvers) that would be needed for replication.
Experiment Setup Yes The time horizon T is set to 10,000, which is enough to observe the performance difference; simulations are repeated 1,000 times, and the number of bootstraps B is set to 500.