Identifiability of Linear AMP Chain Graph Models
Authors: Yuhao Wang, Arnab Bhattacharyya10080-10089
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We also conduct experiments comparing our algorithm s performance against existing baselines. In this section, we compare the performance of Algorithm 1 and Algorithm 2 on synthetic datasets to state-of-the-art methods for AMP chain graph structure learning**. |
| Researcher Affiliation | Academia | Yuhao Wang and Arnab Bhattacharyya National University of Singapore yuhaowang@u.nus.edu, arnabb@nus.edu.sg |
| Pseudocode | Yes | Algorithm 1: Our algorithm for learning the topological order of a chain graph with chain component decomposition D of size t. Algorithm 2: Infinite sample algorithm for learning the topological order of a chain graph with unknown chain components. |
| Open Source Code | Yes | Code is available at https://github.com/YohannaWANG/DCOV |
| Open Datasets | Yes | The ECOLI70 graph provided by (Schafer and Strimmer 2005) contains 46 nodes and 70 edges. The MAGIC-NIAB graph from (Scutari et al. 2014) contains 44 nodes and 66 edges. The MAGIC-IRRI graph contains 64 nodes and 102 edges. |
| Dataset Splits | No | The paper mentions generating synthetic data with a sample size (n = 1000) but does not provide specific percentages or counts for training, validation, or test splits. It also mentions using real datasets from R packages but without detailing splits. |
| Hardware Specification | Yes | The experiments were conducted on an Intel Core i7-9750H 2.60GHz CPU. |
| Software Dependencies | Yes | We implement Algorithm 2 using the Matlab toolbox Submodular Function Optimization (Krause 2010). All the baseline algorithms above are implemented using R-packages (licensed under GPL-2 or GPL-3) such as ggm (Marchetti et al. 2006), pcalg (Kalisch et al. 2012), mgcv (Wood and Wood 2015), np (Racine and Hayfield 2020), and lcd (Ma, Xie, and Geng 2008). We use rpy2 (Gautier 2012) to access R-packages from Python. |
| Experiment Setup | Yes | The observational i.i.d. data Xτ = MτXPa(τ) + Zτ is generated with a sample size n = 1000 and a variable size d {10, 20, 30, 40, 50}. We implement Algorithm 2 using the Matlab toolbox Submodular Function Optimization (Krause 2010). We set the p-value with significance level of 0.001 for determining the parents of the node. |