Gaussian Graphical Modelling Without Independence Assumptions for Uncentered Data
Authors: Bailey Andrew, David R. Westhead, Luisa Cutillo
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We first compare performances on synthetic data, in which the graphs are generated from a Barabasi-Albert (power-law) distribution. We generate means for our synthetic data from a variety of distributions; see Table 1. We also compare them on Erdos-Renyi graphs, achieving similar results, but for conciseness we defer those results to the appendix. Next, we compare performances on the real-world COIL-20 video dataset ((Nene, Nayar, and Murase 1996)), demonstrating clear improvements. Finally, we show that, without properly taking into account the mean, precision matrix estimation will yield clearly wrong results on transcriptomics datasets such as the mouse embryo stem cell E-MTAB-2805 dataset (Buettner et al. 2015). |
| Researcher Affiliation | Academia | Bailey Andrew, David R Westhead, Luisa Cutillo University of Leeds EMAIL, EMAIL, EMAIL |
| Pseudocode | Yes | See Algorithm 1 for a pseudocode presentation of our algorithm. |
| Open Source Code | Yes | Code https://github.com/Bailey Andrew/Noncentral-KSNormal |
| Open Datasets | Yes | Next, we compare performances on the real-world COIL-20 video dataset ((Nene, Nayar, and Murase 1996)), demonstrating clear improvements. Finally, we show that, without properly taking into account the mean, precision matrix estimation will yield clearly wrong results on transcriptomics datasets such as the mouse embryo stem cell E-MTAB-2805 dataset (Buettner et al. 2015). |
| Dataset Splits | No | The paper describes the datasets used (synthetic, COIL-20, E-MTAB-2805) and discusses preprocessing steps or how correct edges were defined for evaluation, but it does not specify explicit training/test/validation splits with percentages, counts, or a detailed methodology for partitioning the data to reproduce experiments. |
| Hardware Specification | Yes | All experiments were run on a 2020 Mac Book Pro with an M1 chip and 8 GB of RAM. |
| Software Dependencies | Yes | Our method was implemented using Num Py 1.25.2 (Harris et al. 2020) and Sci Py 1.12.0 (Virtanen et al. 2020); for precision matrix routines, we used Greenewald, Zhou, and Hero s reference implementation of Tera Lasso (2019) as well as Gm GM 0.5.3. |
| Experiment Setup | Yes | We chose thresholding/regularization parameters such that there would be approximately 144 edges. |