Learning Mixed Latent Tree Models

Authors: Can Zhou, Xiaofei Wang, Jianhua Guo

JMLR 2020 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on the simulated and real data support that our method is valid for mining the hierarchical structure and latent information. In this section, we performed numerical experiments on both simulated and real data sets.
Researcher Affiliation Academia Can Zhou EMAIL Xiaofei Wang EMAIL Jianhua Guo EMAIL KLAS and School of Mathematics and Statistics Northeast Normal University Changchun, China.
Pseudocode Yes Algorithm 1: Structural Learning for Latent Trees (SLLT) Input : Observed variables V and information distances duv for any u, v V; Output: A tree structure T; ... Algorithm 2: Parameter Estimation for Mixed Latent Trees (PEMT) Input : A latent tree with a root, the first order moments EXu for u V, the second order moments EX2 u for u Vc and the moment matrices Euvw and Euv for u, v, w V. Output: All conditional probability matrices on edges in T.
Open Source Code No The paper does not provide an explicit statement about releasing source code for the methodology, nor does it include a direct link to a code repository.
Open Datasets Yes In this part, we applied the SLLT algorithm to a Forest Cover Type dataset, which is available from the University of California, Irvine (UCI) machine learning data set repository. ... For more details about ELU, please visit the following link: https://archive.ics.uci.edu/ml/datasets/Covertype . For more details about basic soil components, please visit the following link: https://casoilresource.lawr.ucdavis.edu/soil_web/ssurgo.php?action=list_mapunits& areasymbol=co645
Dataset Splits No The paper describes generating synthetic data and using a real-world dataset (Forest Cover Type) but does not specify any training, testing, or validation splits for either. The experiments focus on structural learning and parameter estimation consistency rather than performance on predefined data splits.
Hardware Specification Yes All of the experiments were performed using R on a desktop with an Intel Core i5-3470 CPU 3.2 GHz and 16 GB RAM.
Software Dependencies No All of the experiments were performed using R on a desktop with an Intel Core i5-3470 CPU 3.2 GHz and 16 GB RAM. The paper mentions "R" but does not provide specific version numbers for R or any libraries used.
Experiment Setup Yes In the structural learning simulation, we started the value of the threshold ε from 0.1 and let it increase with the step size 0.1 until the SLLT algorithm obtained a tree. We set τ1 = 3 and τ2 = 5. ... We ran the EM with random initialization and 100 iterations.