A Reinforcement Learning Framework for Dynamic Mediation Analysis

Authors: Lin Ge, Jitao Wang, Chengchun Shi, Zhenke Wu, Rui Song

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The superior performance of the proposed method is demonstrated through extensive numerical studies, theoretical results, and an analysis of a mobile health dataset.
Researcher Affiliation Academia 1North Carolina State University 2University of Michigan, Ann Arbor 3London School of Economics and Political Science.
Pseudocode No No pseudocode or algorithm blocks were explicitly labeled or presented in the paper.
Open Source Code Yes A Python implementation of the proposed procedure is available at https://github.com/linlinlin97/Mediation RL.
Open Datasets Yes In this section, we apply the proposed MR estimators to analyze the real dataset from the IHS (Ne Camp et al., 2020)
Dataset Splits Yes It is worth noting that we used cross-validation to estimate the ATE of πopt.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory specifications, or cluster configurations) used for running experiments are provided in the paper.
Software Dependencies No The paper mentions 'A Python implementation' but does not specify version numbers for Python or any other software dependencies, libraries, or solvers used for the experiments.
Experiment Setup Yes We consider a scenario with discrete states, actions, mediators, and rewards. We set time T = 50, and S0 for each trajectory is sampled from a Bernoulli distribution with a mean probability of 0.5. Denote the sigmoid function as expit( ). Following the behavior policy, the action At {0, 1} is sampled from a Bernoulli distribution, where Pr(At = 1|St) = expit(1.0 2.0St). Observing St and At, the mediator Mt {0, 1} is drawn from a Bernoulli distribution with Pr(Mt = 1|St, At) = expit(1.0 1.5St +2.5At).