Estimation of Graphical Models through Structured Norm Minimization
Authors: Davoud Ataee Tarzanagh, George Michailidis
JMLR 2017 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We illustrate the superior performance of the proposed framework on a number of synthetic data sets generated from both random and structured networks. Further, we apply the method to a number of real data sets and discuss the results. |
| Researcher Affiliation | Academia | Davoud Ataee Tarzanagh EMAIL Department of Mathematics UF Informatics Institute University of Florida Gainesville, FL 32611-8105, USA; George Michailidis EMAIL Department of Statistics UF Informatics Institute University of Florida Gainesville, FL 32611-8545, USA |
| Pseudocode | Yes | Algorithm 1 Multi-Block ADMM Algorithm for Solving (19).; Algorithm 2 Non-monotone Barzilai Borwein Method for solving (33) |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. It mentions using third-party packages (CONTEST, UGM) and refers to supplementary materials for a list of genes, but not for the code itself. |
| Open Datasets | Yes | The data set under study considers gene expression profiles of lung cancer tumors... http://www.broadinstitute.org/cgibin/cancer/publications/view/87.; The next example involves a data set 3 containing 1427 documents... http://qwone.com/~jason/20Newsgroups/.; We use monthly stock returns data...obtained from the University of Chicago s Center for Research in Security Prices database (CRSP). |
| Dataset Splits | Yes | We randomly partition the data into 999 training, 214 validation and 214 test examples, corresponding to a 70/15/15 split (Rao et al., 2016). |
| Hardware Specification | Yes | All the algorithms have been implemented in the MATLAB R2015b environment on a PC with a 1.8 GHz processor and 6GB RAM memory. |
| Software Dependencies | Yes | All the algorithms have been implemented in the MATLAB R2015b environment... The CONTEST 1 package is used to generate the synthetic graphs, and the UGM 2 package to implement Gibbs sampling for estimating the Ising Model. |
| Experiment Setup | Yes | The penalty parameters λe and {λi}n i=1 play an important rule for the convex decomposition to be successful. We learn them through numerical experimentation (see Figures 5 and 6) and set them respectively to ϱ = 4, λe = 1, λ1, λ2 = 0.5λe, ˆλi = 0.25λe, and λi+1 = 2λi for i = 2, . . . , n. ...terminated either when Θk Θk 1 2 F / Θk 1 2 F τ, τ = 1e 5, or the number of iterations and CPU times exceed 1,000 and 10 minutes, respectively. |