MixRFb: Mixed Logistic Random Forest for Binary Data

In lschen-stat/MixRF: A Random-Forest-Based Approach for Imputing Clustered Incomplete Data

Description

Mixed Logistic Random Forest for Binary Data

Usage

MixRFb(Y, x, random, data, initialRandomEffects = 0, ErrorTolerance = 0.001,
  MaxIterations = 200, ErrorTolerance0 = 0.001, MaxIterations0 = 15,
  verbose = FALSE)

Arguments

Y	The outcome variable.
x	A formula string contains the predictors.
random	A string in lme4 format indicates the random effect model.
data	The data set as a data frame.
initialRandomEffects	The initial values for random effects.
ErrorTolerance	The tolerance for log-likelihood.
MaxIterations	The maximum iteration times for each run of PQL.
ErrorTolerance0	The tolerance for eta (penalized quasi-likelihood, PQL).
MaxIterations0	The maximum iteration times for PQL.
verbose	The option to monitor each run of PQL or not.

Value

A list contains the random forest, mixed model, and random effects. See the example below for the usage. A predict() function is also available below.

Examples

# example data (http://stats.stackexchange.com/questions/70783/how-to-assess-the-fit-of-a-binomial-glmm-fitted-with-lme4-1-0)
dat <- read.table("http://pastebin.com/raw.php?i=vRy66Bif")

library(party)
library(lme4)

source('MixRFb.r')
system.time(tmp <- MixRFb(Y=dat$true, x='factor(distance) + consequent + factor(direction) + factor(dist)', random='(1|V1)',
                          data=dat, initialRandomEffects=0,
                          ErrorTolerance=1, MaxIterations=200,
                          ErrorTolerance0=0.3, MaxIterations0=15, verbose=T))

# tmp$forest
# tmp$MixedModel
# tmp$RandomEffects

# eta
pred1 = predict.MixRF(tmp, dat, EstimateRE=TRUE)
prob = 1/(1+exp(-pred1))
res = (prob>.5)

# classification
table(res,dat$true)

lschen-stat/MixRF documentation built on May 14, 2019, 11:27 a.m.