Unfolding-Model-Based Visualization: Theory, Method and Applications

Authors: Yunxiao Chen, Zhiliang Ying, Haoran Zhang

JMLR 2021 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Simulation studies and real data examples are presented in Sections 4 and 5, respectively. We illustrate the proposed method through two applications, one on movie rating and the other on senate roll call voting.
Researcher Affiliation Academia Yunxiao Chen EMAIL Department of Statistics London School of Economics and Political Science London, WC2A 2AE, UK Zhiliang Ying EMAIL Department of Statistics Columbia University New York, NY 10027, USA Haoran Zhang EMAIL Shanghai Center for Mathematical Sciences Fudan University Shanghai, 200433, China
Pseudocode Yes Algorithm 1 (Alternating minimization algorithm) Input: Data (yij)N J, pre-specified dimension K+, constraint M, iteration number m = 1, and the initial values θ(0) 1 , ..., θ(0) N and a(0) 1 , ..., a(0) J in RK+. Algorithm 2 (Two-step procedure for bi-cluster analysis) Step 1: Apply Algorithm 1 and obtain estimates {ˆθ1, ..., ˆθN, ˆa1, ..., ˆa J}.
Open Source Code Yes 1. An R package has been developed and can be downloaded from https://github.com/hrzhang16/mmdu.
Open Datasets Yes The movie dataset is a subset from the famous Movie Lens dataset (Harper and Konstan, 2016). The dataset can be downloaded from https://grouplens.org/datasets/movielens/100k/. This dataset contains the voting records from 100 senators to 675 roll calls in years 2003 and 20044. The dataset can be downloaded from https://legacy.voteview.com/dwnl.htm.
Dataset Splits No The paper describes using a subset of the Movie Lens dataset containing 943 users and 338 movies, and a senate roll call voting dataset from the 108th congress involving 100 senators and 675 roll calls. It mentions generating 100 independent datasets for simulation studies. However, it does not specify any training, validation, or test splits for these datasets, nor does it refer to standard predefined splits.
Hardware Specification Yes 2. All the computation is conducted on a single Intel Gold 6130 core.
Software Dependencies No The paper mentions "our implementation of Algorithm 1 in statistical software R" and "An R package has been developed and can be downloaded from https://github.com/hrzhang16/mmdu." However, no specific version numbers for R or any other software packages or libraries are provided.
Experiment Setup Yes In the application of Algorithm 1, the constraint M is set to 1.5. using ten random starting points and then choosing the result that gives the largest likelihood function value. The results presented below are based on the choice δ = 0.1, but we point out that other choices of δ (δ = 0.05, 0.15, 0.2) have also been tried which all lead to very similar results. The constraint constant M is set to 3.5 when applying Algorithm 1.