Learning Adjustment Sets from Observational and Limited Experimental Data

Authors: Sofia Triantafillou, Greg Cooper9940-9948

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments, Simulations setup. We examined the performance of our method in three different settings:, Evaluation measures. We examined the performance of our algorithms in terms of their ability to improve causal effect estimation for the observational population:, Results: Fig 3 shows results for random SMCMs (top row), and for SMCMs where the covariates are known to be pre-treatment. FAS improves the estimation of PX(Y )(| | closer to zero, lower variance) compared to Dexp, particularly in cases in where the experimental data come from a selected population.
Researcher Affiliation Academia So fia Trianta fillou, Greg Cooper University of Pittsburgh, Department of Biomedical Informatics {sot16, gfc}@pitt.edu
Pseudocode Yes Algorithm 1: Find Adjustment Set (FAS), Algorithm 2: Score Exp, Algorithm 3: Selection BN
Open Source Code No The paper does not include any statement about releasing source code for the described methodology, nor does it provide a link to a code repository.
Open Datasets Yes We used RCT data from the STOMP trial (Parker et al. 2013), which estimated the effect of statin use on myalgia.
Dataset Splits No The paper mentions using 'Observational data Dobs' and 'Experimental data Dexp' and describes their sample sizes (e.g., '10,000 samples' for Dobs), but it does not provide specific train/validation/test splits, percentages, or sample counts, nor does it reference standard predefined splits for these datasets.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory, or specific computing environments) used to run its simulations or analyze the data.
Software Dependencies No The paper states 'For Learn BN, we used FGES, an optimized version of GES (Chickering 2002)', which names a software, but it does not provide a specific version number for FGES/GES or any other software dependencies.
Experiment Setup Yes We used niters=100. We simulated Dobs with 10,000 samples from DAGs with mean in-degree 2. Each DAG includes a pair X, Y where X causes Y , and 10 additional covariates: 6 observed and 4 latent. We used two types of DAGs: (i) random DAGs and (ii) DAGs where all the additional covariates are pre-treatment. Variables were discrete with 2-3 categories each and random parameter values P(X|Pa(X)). Selection bias was imposed by adding binary selection nodes Si and random parameters P(Si=1|Vi).