Constraint-based Causal Structure Learning with Consistent Separating Sets
Authors: Honghao Li, Vincent Cabeli, Nadir Sella, Herve Isambert
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conducted a series of benchmark structure learning simulations to study the differences between the original PC-stable algorithm and the proposed modifications ensuring consistent separating sets. For each simulation setting, we first quantified the fraction of inconsistent separating sets predicted by the original PC-stable algorithm, Figure 3. We then compared the performance of the original PC-stable (algorithm 1 and algorithm 2), orientation-consistent PC-stable (algorithm 3) and skeletonconsistent PC-stable (algorithm 4), for different significance levels α, in terms of the precision and recall of the adjacencies found in the inferred graph with respect to the true skeleton, Figures 4 and 5. |
| Researcher Affiliation | Academia | Honghao Li, Vincent Cabeli, Nadir Sella, Hervé Isambert Institut Curie, PSL Research University, CNRS UMR168, Paris {honghao.li, vincent.cabeli, nadir.sella, herve.isambert}@curie.fr |
| Pseudocode | Yes | Algorithm 1 The PC Algorithm, Algorithm 2 Find skeleton and separating sets (Step 1 of PC-stable algorithm), Algorithm 3 Sepset consistent PC algorithm (1st version, orientation consistency), Algorithm 4 Sepset consistent PC algorithm (2nd version, skeleton consistency) |
| Open Source Code | No | The paper mentions 'Reconstruction benchmarks were performed with pcalg s (Kalisch et al., 2012) PC-stable implementation, modified for enforcing separating set consistency'. However, it does not provide an explicit statement or link for the open-sourcing of their modifications or code. |
| Open Datasets | Yes | The data-sets used for the numerical experiments were generated with the following scheme. The underlying DAGs were generated with TETRAD (Scheines et al., 1998) as scale-free DAGs with 50 nodes (α = 0.05, β = 0.4, average total degree d(G) = 1.6) using a preferential attachment model and orienting its edges based on a random topological ordering of the vertices. In addition, we also generated data-sets for the classical benchmarks Insurance (27 nodes, 52 links, 984 paramaters), Hepar2 (70 nodes, 123 links, 1453 paramaters) and Barley (48 nodes, 84 links, 114005 paramaters) networks from the Bayesian Network repository (Scutari, 2010). |
| Dataset Splits | No | The paper describes data generation schemes and sample sizes (e.g., 'N=500 samples', 'N=1000 samples') and the number of networks used (100), but it does not specify explicit training, validation, or test dataset splits (e.g., 80/10/10 split, or exact sample counts for each split). |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models, memory, or types of computing instances used for the experiments. |
| Software Dependencies | No | The paper mentions 'Reconstruction benchmarks were performed with pcalg s (Kalisch et al., 2012) PC-stable implementation' and 'bnlearn R Package (Scutari, 2010)' but does not specify their version numbers or other software dependencies with version numbers. |
| Experiment Setup | Yes | The (conditional) independence test used in all simulations is a linear (partial) correlation with Fisher s z-transformation. Data-sets were simulated with linear structural equation models for three settings : strong, medium and weak interactions (with respective coefficient ranges [0.2, 0.7], [0.1, 0.5], and [0, 0.3] and covariance ranges [0.5, 1.5], [0.5, 1], and [0.2, 0.7]). For each simulation setting, we first quantified the fraction of inconsistent separating sets predicted by the original PC-stable algorithm, Figure 3. We then compared the performance of the original PC-stable (algorithm 1 and algorithm 2), orientation-consistent PC-stable (algorithm 3) and skeletonconsistent PC-stable (algorithm 4), for different significance levels α. |