Lokālie lineārie un vispārinātie lineārie meži
Loading...
Date
Authors
Sjomkāns, Oskars
Advisor
Journal Title
Journal ISSN
Volume Title
Publisher
Latvijas Universitāte
Language
lav
Abstract
Darbs koncentrējas uz mēģinājumu uzlabot LLF svaru aprēķinam izmantoto pieeju, kas balstās uz regresijas mežu izveidi. Veidojot koku random regression forest ansamblī, eksistējošo pieeju ar kvadrātisko starpību summu izvērtējamā šķēluma abās pusēs aizvieto ar parametriska, vispārinātā lineārā modeļa ticamībām abās izvērtējamā šķēluma pusēs.
This thesis focuses on an attempt to improve the approach that LLF uses to compute weights which are based on random regression forests. When building a tree in a random forest, the existing approach that uses sum of quadratic differences on both sides of a potential split is replaced with the likelihoods of a parametric, generalized linear model.
This thesis focuses on an attempt to improve the approach that LLF uses to compute weights which are based on random regression forests. When building a tree in a random forest, the existing approach that uses sum of quadratic differences on both sides of a potential split is replaced with the likelihoods of a parametric, generalized linear model.