Characterizing and Learning Equivalence Classes of Causal DAGs under Interventions

Authors: Karren Yang, Abigail Katcoff, Caroline Uhler

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

Reproducibility Variable Result LLM Response
Research Type Experimental We also propose the first provably consistent algorithm for learning DAGs in this setting and evaluate our algorithm on simulated and biological datasets.
Researcher Affiliation Academia Massachusetts Institute of Technology, Cambridge, MA. Correspondence to: Caroline Uhler <cuhler@mit.edu>.
Pseudocode Yes Algorithm 1 IGSP for general interventions
Open Source Code No The paper does not provide an explicit statement or link for the open-sourcing of the code for the methodology described.
Open Datasets Yes Protein Expression Dataset: We evaluated our algorithm on the task of learning a protein network from a protein mass spectroscopy dataset (Sachs et al., 2005). Gene Expression Dataset: We also evaluated IGSP on a single-cell gene expression dataset (Dixit et al., 2016).
Dataset Splits No The paper describes using datasets for evaluation, but does not explicitly provide specific train/validation/test dataset splits, percentages, or cross-validation details for reproducibility.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details, such as library or solver names with version numbers, needed to replicate the experiment.
Experiment Setup Yes For each simulation, we sampled 100 DAGs from an Erd os Renyi random graph model with an average neighborhood size of 1.5 and p {10, 20} nodes. The data for each causal DAG G was generated using a linear structural equation model with independent Gaussian noise: X = AX + ϵ, where A is an upper-triangular matrix with edge weights Aij = 0 if and only if i j, and ϵ N(0, Ip). For Aij = 0, the edge weights were sampled uniformly from [ 1, 0.25] [0.25, 1]. We simulated perfect interventions on i by setting the column A,i = 0; inhibiting interventions by decreasing A,i by a factor of 10; and imperfect interventions with a success rate of α = 0.5.