MoE-mph-method: Fit method for mph/miph class, using mixture-of-experts...

In matrixdist: Statistics for Matrix Distributions

Fit method for mph/miph class, using mixture-of-experts regression

Description

Fit method for mph/miph class, using mixture-of-experts regression

Usage

## S4 method for signature 'mph'
MoE(
  x,
  formula,
  y,
  data,
  alpha_mat = NULL,
  delta = numeric(0),
  stepsEM = 1000,
  r = 1,
  maxit = 100,
  reltol = 1e-08,
  rand_init = T
)

Arguments

x	An object of class mph.
formula	a regression formula.
y	A matrix of observations.
data	A data frame of covariates (they need to be scaled for the regression).
alpha_mat	Matrix with initial distribution vectors for each row of observations.
delta	Matrix with right-censoring indicators (1 uncensored, 0 right censored).
stepsEM	Number of EM steps to be performed.
r	Sub-sampling parameter, defaults to 1 (not supported for this method).
maxit	Maximum number of iterations when optimizing the g function (inhomogeneous likelihood).
reltol	Relative tolerance when optimizing g function.
rand_init	Random initiation in the R-step of the EM algorithm.

Examples

under_mph <- mph(structure = c("general", "general"), dimension = 3) 
x <-  miph(under_mph, gfun = c("weibull", "weibull"), gfun_pars = list(c(2), c(3)))
n <- 100
responses <- cbind(rexp(n), rweibull(n, 2, 3))
covariates <- data.frame(age = sample(18:65, n, replace = TRUE) / 100, income = runif(n, 0, 0.99))
f <- responses ~ age + income
MoE(x = x, formula = f, y = responses, data = covariates, stepsEM = 20)

matrixdist documentation built on Aug. 8, 2023, 5:06 p.m.