###############################################################################
###############################################################################
# Chapter 8 Supplemental
###############################################################################
###############################################################################

# Example 5 Using Random Forests

# Random Forests - Application to the Chiaretti (2004) data. With respect to...
library(randomForest)
library("hgu95av2.db"); library(ALL); data(ALL)
ALLB123 <- ALL[,ALL$BT %in% c("B1","B2","B3")]
pano <- apply(exprs(ALLB123), 1, function(x) anova(lm(x ~ ALLB123$BT))$Pr[1])
names <- featureNames(ALL)[pano<0.000001]
symb <- mget(names, env = hgu95av2SYMBOL)
ALLBTnames <- ALLB123[names, ]
probedat <- as.matrix(exprs(ALLBTnames))
row.names(probedat)<-unlist(symb)
diagnosed <- factor(ALLBTnames$BT)

rForest <- randomForest(diagnosed ~ ., data = data.frame(t(probedat)), ntree=1000, norm.votes=FALSE)
print(rForest)
importance(rForest)
plot(rForest)
legend("topright", legend=colnames(rForest$err.rate), col=1:4, pch=19)

# Now try multiple forests recursively. Notice that no two random foests are exactly the same
# Can actually create multiple random forests and combine them

rf1 <- randomForest(diagnosed ~ ., data = data.frame(t(probedat)), ntree=1000, norm.votes=FALSE)
rf2 <- randomForest(diagnosed ~ ., data = data.frame(t(probedat)), ntree=1000, norm.votes=FALSE)
rf3 <- randomForest(diagnosed ~ ., data = data.frame(t(probedat)), ntree=1000, norm.votes=FALSE)
rf.all <- randomForest::combine(rf1, rf2, rf3)
print(rf.all)