Determining conditional Gaussian distributions for decomposable graphical models: - A new method
Author
Lund, Peter Enemark
Term
4. term
Education
Publication year
2016
Pages
37
Abstract
Dette speciale introducerer en ny metode til at beregne betingede Gaussiske fordelinger i dekomponerbare modeller—modeller der kan opdeles i mindre dele, hvilket gør beregninger mere håndterbare. Betingede Gaussiske fordelinger beskriver, hvordan nogle variabler opfører sig givet værdierne af andre. For at gøre metoden tilgængelig gennemgås først grundlæggende grafteori, den multivariate normalfordeling og maksimal sandsynlighedsestimation. Derefter præsenteres den nye metode sammen med den generelle formulering for dekomponerbare modeller. Til sidst forklares implementeringen i R, og metoden sammenlignes med en enklere løsning.
This thesis introduces a new method for computing conditional Gaussian distributions in decomposable models—models that can be broken into smaller parts to make calculations more manageable. Conditional Gaussian distributions describe how some variables behave given the values of others. To make the method accessible, the thesis first reviews basic graph theory, the multivariate normal distribution, and maximum likelihood estimation. It then presents the new method along with the general case for decomposable models. Finally, it explains how to implement the method in R and compares it with a simpler approach.
[This abstract was generated with the help of AI]
Documents