Causal Discovery in Heterogeneous Environments Under the Sparse Mechanism Shift Hypothesis

Authors: Ronan Perry, Julius von Kügelgen, Bernhard Schölkopf

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, we verify behavior predicted by the theory and compare multiple estimators and score functions to identify the best approaches in practice.
Researcher Affiliation Academia 1Max Planck Institute for Intelligent Systems, Tübingen, Germany 2University of Cambridge, United Kingdom
Pseudocode No The paper describes the Mechanism Shift Score and its estimand, but it does not include a formally labeled 'Pseudocode' or 'Algorithm' block with structured steps.
Open Source Code Yes All code and experiments are available at https://github.com/rflperry/sparse_shift
Open Datasets No The paper describes generating synthetic data using an Erd os-Rènyi model and mentions a
Dataset Splits No The paper mentions using random DAGs over variables with specific edge densities, number of environments, and shift fractions. It also notes using
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used to run the experiments.
Software Dependencies No The paper mentions the use of specific software components like "KCI and Fisher-Z are implemented by the causal-learn package [GNU General Public License]" and "causaldag package [3-Clause BSD license]". However, it does not provide specific version numbers for these packages or any other software dependencies, which is necessary for reproducibility.
Experiment Setup Yes Random DAGs are sampled using an Erd os-Rènyi model [12] in which each edge has some fixed probability of existing (the edge density). In each environment, a random set of variables experience a mechanism change according to a fixed number or fraction of shifts. Each variable j in environment e has a randomly sampled mechanism... three environments are sampled, with 500 samples and two mechanism shifts per environment.