Package of 'glm.tree'

Change Log:

• 1.2; 3rd August 2012. Define namespace
• 31th March 2011. Debugged because older version can't conduct tree-glm only within a specific range of longitude
• 17th Dec 2010. Add binary zip package for Windows XP

Download and Install:
[source code] glm.tree_1.2.tar.gz

• Install the 'tar.gz' ball by R command such as

   
  command line> R CMD INSTALL glm.tree_1.0.tar.gz
  

• GO to menu -> package -> Install zip file in local directory -> Select the zip file

Required additional package: biglm

Usage: See help after calling the library of 'glm.tree' such as

 
library(glm.tree)  
help(glm.tree)
 

Examples: (same as shown in help of glm.tree)

library(glm.tree)
data(ldata)
# install.packates(biglm) # if 'biglm' packages are not installed.  
library(biglm)
ldata$lcpue <- log(ldata$N)
ldata$area <- 0
# Dependent variable should be in 'ldata$lcpue', independent
# variable of longitude and latitude should be in 'ldata$lon' and
# ldata$lat, respectively
str(ldata)
# remove zero catch data
ldata <- subset(ldata,N>0)

# create area stratification

# conduct iteration until obtaining minimum AIC
# ==> !! caution: this takes too much time and computer memory
res0 <- make.treeglm(expression(lcpue~as.factor(year)+as.factor(area)),
                          data=ldata,debug.mode=F,max.split=Inf,graph=F,IC="AIC")
# conduct only 10th iterations
res <- make.treeglm(expression(lcpue~as.factor(year)+as.factor(area)),
                          data=ldata,debug.mode=F,max.split=10,graph=F,IC="AIC")
plotworld(res$data,area=res$area[[9]])
plot(res$summary.stat$Num.area,res$summary.stat$AIC)

# compare results
ldata$area2 <- res$area[[2]]
ldata$area9 <- res$area[[9]]
ldata$area14 <- res$area[[14]]
res.glm2 <- glm(lcpue~as.factor(year)+as.factor(area2),data=ldata)
res.glm9 <- glm(lcpue~as.factor(year)+as.factor(area9),data=ldata)
res.glm14 <- glm(lcpue~as.factor(year)+as.factor(area14),data=ldata)

par(mfrow=c(2,2))
plot(exp(c(res.glm2$coef[1],res.glm2$coef[1]+res.glm2$coef[2:16])),
    type="b",main="Iteration=3")
plot(exp(c(res.glm9$coef[1],res.glm9$coef[1]+res.glm9$coef[2:16])),
    type="b",main="Iteration=9")
plot(exp(c(res.glm14$coef[1],res.glm14$coef[1]+res.glm14$coef[2:16])),
    type="b",main="Iteration=14") # Closer to true abundance trends

# Another option to use BIC as an information criteria
res.bic <- make.treeglm(expression(lcpue~as.factor(year)+as.factor(area)),
                          data=ldata,debug.mode=F,max.split=15,graph=F,IC="BIC")
# Another option to create area stratification,
#     for avoiding area strata with missing data
res.md <- make.treeglm(expression(lcpue~as.factor(year)+as.factor(area)),
                          data=ldata,debug.mode=F,max.split=15,graph=F,IC="MD",IC2="AIC")

Past packages:

• glm.tree_1.1.tar.gz, glm.tree_1.1.zip
• glm.tree_1.0.tar.gz, glm.tree_1.0.zip