Sample Efficient Bayesian Learning of Causal Graphs from Interventions
Authors: Zihan Zhou, Muhammad Qasim Elahi, Murat Kocaoglu
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We compare our algorithm against various baseline methods on simulated datasets, demonstrating its superior accuracy measured by the structural Hamming distance between the learned DAG and the ground truth.Additionally, we present a case study showing how this algorithm could be modified to answer more general causal questions without learning the whole graph. As an example, we illustrate that our method can be used to estimate the causal effect of a variable that cannot be intervened. |
| Researcher Affiliation | Academia | Zihan Zhou* Electrical and Computer Engineering Purdue University zhou1248@purdue.edu Muhammad Qasim Elahi* Electrical and Computer Engineering Purdue University elahi0@purdue.edu Murat Kocaoglu Electrical and Computer Engineering Purdue University mkocaoglu@purdue.edu |
| Pseudocode | Yes | Algorithm 1: Sample efficient causal discovery in Bayesian approach Algorithm 2: G-separating system of a given unoriented graph G Algorithm 3: Enumerating all interventional distributions and priors of an intervention set I on an undirected graph G |
| Open Source Code | Yes | The algorithm code is provided at https://github.com/Causal ML-Lab/ Bayesian_Sample Efficient_Discovery. |
| Open Datasets | No | We generate random connected moral DAGs with order n and density ρ using a modified Erd os Rényi sample approach, similar to the process in Squires et al. [2020]. The paper describes how the simulated datasets are generated but does not provide concrete access information (link, DOI, specific repository, or citation to a public dataset). |
| Dataset Splits | No | The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce the data partitioning. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper mentions 'Chi-Square independence test from the Causal Discovery Toolbox [Kalainathan et al., 2020]' but does not provide specific ancillary software details with version numbers needed to replicate the experiment. |
| Experiment Setup | Yes | We generate random connected moral DAGs with order n and density ρ using a modified Erd os Rényi sample approach... For each setting of n and ρ, we sample 50 random DAGs and calculate the mean and standard deviation of SHD by each causal discovery algorithm. For AVICI, we fine-tuned the pretrained models, scm-v0 and neurips-rff, with 50 random complete graphs, each with 1000 observational samples. At each step for both methods, we choose the interventional target randomly. |