Homogeneity Structure Learning in Large-scale Panel Data with Heavy-tailed Errors
Authors: Di Xiao, Yuan Ke, Runze Li
JMLR 2021 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive numerical experiments demonstrate the advantage of the proposed learning procedure over conventional methods especially when the data are generated from heavy-tailed distributions. |
| Researcher Affiliation | Academia | Di Xiao EMAIL Department of Statistics University of Georgia Athens, GA 30602, USA Yuan Ke EMAIL Department of Statistics University of Georgia Athens, GA 30602, USA Runze Li EMAIL Department of Statistics The Pennsylvania State University University Park, PA 16802, USA |
| Pseudocode | Yes | Algorithm 1 Change-points detection with wild binary segmentation Algorithm 2 Robust homogeneity structure learning |
| Open Source Code | No | The WBS package is available at https://cran.r-project.org/web/packages/wbs/index.html. The Farm Test package is available at https://cran.r-project.org/web/packages/Farm Test/index.html. These are third-party tools used in the paper, not the authors' own implementation code for the methodology described. |
| Open Datasets | Yes | A detailed description of this panel data can be found in Appendix A of Ludvigson and Ng (2009). 5. The data set is available at https://www.epa.gov/outdoor-air-quality-data. |
| Dataset Splits | Yes | Start from the first day in the data set, we use a window size of 250 days as the training set to predict the next 50 days. Each time, the window moves 20 days forward. |
| Hardware Specification | Yes | The average wall-time running costs of Our, Ah and Mom are 0.8, 11 and 7 seconds per replication, respectively. For each method, we simulate 200 replications on the same computer cluster node with Intel(R) Xeon(R) CPU E5-2690 v3 @ 2.60GHz and 256Gb RAM. |
| Software Dependencies | No | The WBS package is available at https://cran.r-project.org/web/packages/wbs/index.html. The Farm Test package is available at https://cran.r-project.org/web/packages/Farm Test/index.html. The paper mentions these R packages but does not provide specific version numbers for them or the R environment itself. |
| Experiment Setup | Yes | Throughout this section, we set N = 100, T = 200, p = 30 and q = 2. For each scenario, we simulate 200 replications unless otherwise specified. The structures (i) and (ii) have 5 and 9 groups, respectively. For both structures, the signal strength r is set to be 1 (week), 2 (medium), or 4 (strong). In our numerical studies, we choose R = 5000 and ξ = 2 ln T which are the default values in the WBS package. With qmax = 4 and CT = 0.01, the modified ratio method (13) estimates the number of factors to be 1. we propose to use the 5-fold cross-validation to select their robustification parameters in our numerical studies. |