The Potts-Ising model for discrete multivariate data

Authors: Zahra Razaee, Arash Amini

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the ability of the model to capture multivariate dependencies in real data by comparing with existing approaches.
Researcher Affiliation Collaboration Zahra S. Razaee Biostatistics and Bioinformatics Research Center, Cedars Sinai zahra.razaee@cshs.org Arash A. Amini Department of Statistics University of California, Los Angeles aaamini@ucla.edu
Pseudocode No The paper describes the model fitting process using mathematical equations and prose, but it does not include any structured pseudocode or algorithm blocks.
Open Source Code Yes The code for these simulations is available at Git Hub repository aaamini/pois_comparisons.
Open Datasets Yes We consider the following publicly available datasets: Movie Lens 100K Dataset [23] of movie ratings (n = 1037, sparsity 97.8%), and an Amazon customer ratings dataset [24] (n = 745, sparsity 94.3%).
Dataset Splits Yes In the simulations, we split the data into a training and a test set. (...) The three methods POIS, T-PGM, and MIXPOI have hyper-parameters that we tuned by splitting the training set further into training and validation sets (at 70/30 ratio) and evaluating the performance on the validation set using the pair-complement MMD.
Hardware Specification No The paper does not specify any particular hardware components such as CPU or GPU models, or cloud computing instance types used for running the experiments.
Software Dependencies Yes Firth s approach is equivalent to putting a Jeffrey s prior on γi and we use the implementation available in logistf R package [16]. (...) we take advantage of the efficiency of glmnet in using warm starts to compute the entire regularization path as of function of λ. We can then use cross-validation to tune λ as will be discussed in Section 4. [16] G. Heinze et al. R Package logistf . http://CRAN.R-project.org/package=logistf. 2018. [17] J. Friedman et al. R Package glmnet . http://CRAN.R-project.org/package=glmnet. 2018.
Experiment Setup Yes For the POIS model, we used a single regularization parameter λ for the ℓ1 penalty in all the neighborhood regressions. We used 15 values of λ between 10 4 to 10 1.3, equally spaced on the log-scale. (...) The sampling parameter is generally taken to be m = 1000 in our simulations. (...) The results are further averaged over n CV = 5 random training-test splits.