Estimating Identifiable Causal Effects on Markov Equivalence Class through Double Machine Learning

Authors: Yonghan Jung, Jin Tian, Elias Bareinboim

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

Reproducibility Variable Result LLM Response
Research Type Experimental Simulation results corroborate with the theory. and Experimental studies corroborate with the theory. and 6. Experiments We evaluate DML-IDP for estimating Px(y) in Fig. (2a,2b,1). We specify an SCM M for each PAG and generate datasets D from M.
Researcher Affiliation Academia 1Department of Computer Science, Purdue University, USA 2Department of Computer Science, Iowa State University, USA 3Department of Computer Science, Columbia University, USA.
Pseudocode Yes Algorithm 1 IFP(x, y, G(V), P)
Open Source Code No The paper does not provide any explicit statement or link regarding the public availability of its source code.
Open Datasets No We specify an SCM M for each PAG and generate datasets D from M. and All values in the following tables denote parameters for beta distributions, and we sample from each variable according to the parameters. (No concrete access information for a publicly available or open dataset is provided.)
Dataset Splits No We generate 100 datasets for each sample size N. (No specific dataset splits like percentages or absolute counts for training, validation, and test sets are mentioned.)
Hardware Specification No We use a single CPU core for XGBoost training and prediction. (No specific hardware details like CPU model, GPU model, or memory specifications are provided.)
Software Dependencies No Nuisance functions are estimated using standard techniques available in the literature (refer to Appendix C for details), e.g., conditional probabilities are estimated using a gradient boosting model XGBoost (Chen & Guestrin, 2016) and The nuisance functions are estimated with XGBoost (Chen & Guestrin, 2016). (XGBoost is mentioned, but no specific version number is provided.)
Experiment Setup Yes XGBoost is set with default parameters with max_depth=6, subsample=0.8, colsample_bytree=0.8, and n_estimators=100. We use a single CPU core for XGBoost training and prediction. Each experiment is repeated 100 times, and we compute the average AAE.