In sparktseung/LRMoE: Logit-Weighted Reduced Mixture-of-Experts

knitr::opts_chunk$set(echo = TRUE, cache = TRUE)
library(LRMoE)
library(reshape2)
library(matrixStats)

After fitting an LRMoE model, the following predictive functions provide further insights into the dataset. These functions start with predict_ followed by a quantity of interest (e.g. mean_) listed below.

• class: latent class probabilities and the most likely latent class;
• mean: mean of response;
• var: variance of response;
• limit: limited expected value (LEV) of response, that is, $E[{min}(Y, d)]$;
• excess: expected excess value of response, that is, $E[{max}(Y-d, 0)]$; and
• VaRCTE: quantile (or Value-at-Risk/VaR) and conditional tail expectation (CTE, or tail-VaR/TVaR) of response.

These quantities can be calculated based on either the prior and posterior latent class probabilities, as indicated by the suffix of these functions.

• prior: the latent class probabilities are based on the covariates X and logit regression coefficients alpha.
• posterior: the latent class probabilities are based on the covariates X, logit regression coefficients alpha and observed values Y.

The differences of these probabilities can be found in Fung et al. (2019).

The following contains a detailed description of all predictive functions included in the package. Throughout this page, Y is a matrix of response, X a matrix of covariates, alpha a matrix of logit regression coefficients and model a matrix of expert functions.

predict_class_prior
predict_class_posterior
predict_mean_prior
predict_mean_posterior
predict_var_prior
predict_var_posterior
predict_limit_prior
predict_limit_posterior
predict_excess_prior
predict_excess_posterior
predict_VaRCTE_prior
predict_VaRCTE_posterior

sparktseung/LRMoE documentation built on March 21, 2022, 3:22 a.m.