Discrete statistical models with rational maximum likelihood estimator

DOI10.3150/20-BEJ1231MaRDI QIDQ2214240zbMATH OpenFDO

Authors Eliana Duarte, Orlando Marigliano, Bernd Sturmfels

Publication date 7 December 2020

Published in Bernoulli (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1903.06110, https://projecteuclid.org/euclid.bj/1605841239

zbMATH Keywords

algebraic statisticsdiscrete statistical modelsgraphical modelslikelihood geometrymaximum likelihood estimatorreal algebraic geometry

Mathematics Subject Classification ID

Semialgebraic sets and related spaces (14P10) Probabilistic graphical models (62H22) Algebraic statistics (62R01)

Abstract: A discrete statistical model is a subset of a probability simplex. Its maximum likelihood estimator (MLE) is a retraction from that simplex onto the model. We characterize all models for which this retraction is a rational function. This is a contribution via real algebraic geometry which rests on results due to Huh and Kapranov on Horn uniformization. We present an algorithm for constructing models with rational MLE, and we demonstrate it on a range of instances. Our focus lies on models familiar to statisticians, like Bayesian networks, decomposable graphical models, and staged trees.

Review text

This paper is a study on the discrete statistical models. The authors show a look at algebraic statistics to build the maximum likelihood estimator (MLE). They present a theorem that characterizes all discrete statistical models whose MLE is a rational function. They examine models with rational MLE such as decomposable graphical models and Bayesian networks. The authors present an algorithm for constructing models with rational MLE, and they discuss its implementation and some experiments. The input is an integer matrix, and the output is a list of models derived from that matrix. Their results suggest that only a very small fraction of Huh's varieties are statistical models.

Recommendations

• Maximum likelihood for dual varieties
• Bounding the maximum likelihood degree
• Toric invariant theory for maximum likelihood estimation in log-linear models
• Solving the likelihood equations
• Maximum likelihood geometry in the presence of data zeros

Cites work

• A characterization of A-discriminantal hypersurfaces in terms of logarithmic Gauss map
• Chain event graphs
• Conditional independence and chain event graphs
• Equations defining probability tree models
• Equivalence classes of staged trees
• scientific article; zbMATH DE number 575960 (Why is no real title available?)
• scientific article; zbMATH DE number 1134987 (Why is no real title available?)
• Information geometry
• Lectures on algebraic statistics
• Likelihood geometry
• Linear precision for parametric patches
• Moment maps, strict linear precision, and maximum likelihood degree one
• Toric surface patches
• Varieties with maximum likelihood degree one

Cited in

(23)

• Discrete max-linear Bayesian networks
• Staged tree models with toric structure
• Exact P-values for discrete models obtained by estimation and maximization
• Optimal Transport to a Variety
• DiscreteStatisticalModelsWithRationalMLE
• Algebraic structures in statistical methodology. Abstracts from the workshop held December 4--10, 2022
• Nonlinear algebra and applications
• Toric fiber products in geometric modeling
• Families of polytopes with rational linear precision in higher dimensions
• Beyond linear algebra
• Decomposable context-specific models
• Classical iterative proportional scaling of log-linear models with rational maximum likelihood estimator
• Differential equations for Gaussian statistical models with rational maximum likelihood estimator
• D-modules and holonomic functions
• Combinatorial geometry of rational quasi-independence models as toric fiber products
• Direct sampling from conditional distributions by sequential maximum likelihood estimations
• Minimal kinematics on \(\mathcal{M}_{0,n}\)
• Maximum likelihood estimation of log-affine models using detailed-balanced reaction networks
• Classifying one-dimensional discrete models with maximum likelihood degree one
• Maximum likelihood degree of the -stochastic blockmodel
• Representation of context-specific causal models with observational and interventional data
• One-dimensional discrete models of maximum likelihood degree one
• Algorithm for direct sampling from conditional distributions of toric models

Describes a project that uses

Uses Software

• Macaulay2

This page was built for publication: Discrete statistical models with rational maximum likelihood estimator

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2214240)