安裝
install.packages("randomForest", lib = ".", destdir = ".")
                                 ^^^^^^^^^  ^^^^^^^^^^^^^
                                 安裝目錄     下載目錄
install.packages("fastICA", lib = ".", destdir = ".")
install.packages("SparseM", lib = ".", destdir = ".")
install.packages("e1071", lib = ".", destdir = ".")


==============RandomForest用法==========
每次執行時
.libPaths() 
.libPaths(path.expand(".")) 
library(randomForest) 
train <- read.table("ada_train.data")
label <- read.table("ada_train.labels")
valid <- read.table("ada_valid.data")
train_x <- as.matrix(train)
train_y <- as.matrix(label)
set.seed(1)
model <- randomForest(train_x, as.factor(train_y), mtry = 8, ntree = 499, classwt = c(1, 2.62))
model
predict(model, as.matrix(valid))
write(predict(model, as.matrix(valid)),"ada_valid.resu")

# set.seed(1) is used before using randomForest each time
# set seed = 1, ntree = 495~499, 501, mtry = 8, classwt = c(1, 2.62), can get
# 0.1577 in WCCI contest
==============對  ICA  作圖=============
read.matrix.csr("train.csv.out.scale") -> ada
as.matrix(ada$x)-> ada.data
ada.ica <- fastICA(ada.data, n.comp=40)
plot(ada.ica$S,pch=21,bg=c(rainbow(4))[ada$y])
plot(ada.ica$S[,1:2],pch=21,bg=c(rainbow(4))[ada$y])
ada.ica <- fastICA(ada.data, n.comp=5)
plot(ada.ica$S[,1:2],pch=21,bg=c(rainbow(4))[ada$y])

============SVM 格式 作fastICA==========
 
.libPaths() 
.libPaths(path.expand(".")) 
library(e1071)
library(SparseM)
library(fastICA)

read.matrix.csr("train.csv.out.scale") -> ada
as.matrix(ada$x)-> ada.data
ada.ica <- fastICA(ada.data, n.comp=30)
val.ica <- fastICA(val.data, n.comp=30)
val.ica$X%*%ada.ica$K%*%ada.ica$W -> val.dataica
write.matrix.csr(as.matrix.csr(val.dataica), y=val$y, file="ada.ica30.val")
write.matrix.csr(as.matrix.csr(ada.ica$S), y=ada$y, file="ada.ica30.train")
ada.ica <- fastICA(ada.data, n.comp=25)
val.ica <- fastICA(val.data, n.comp=25)
val.ica$X%*%ada.ica$K%*%ada.ica$W -> val.dataica
write.matrix.csr(as.matrix.csr(ada.ica$S), y=ada$y, file="ada.ica25.train")
write.matrix.csr(as.matrix.csr(val.dataica), y=val$y, file="ada.ica25.val")

==========================================
 
ada.ica <- fastICA(ada.data, n.comp=30, method="C")
 
===========RAW DATA格式 作fastICA=========

.libPaths() 
.libPaths(path.expand(".")) 
library(e1071)
library(SparseM)
library(fastICA)
ls()                  #列出舊有物件
dir()                 #列出工作目錄下檔案
rm(list=ls(all=TRUE)) #刪除舊有物件
	
sylva <- read.table("sylva_train.data")
sylva.val = read.table("sylva_valid.data")
sylva.valm = as.matrix(sylva.val)
sylvam = as.matrix(sylva)
sylva.ica <- fastICA(sylvam, n.comp=30)
sylva.val.ica <- fastICA(sylva.valm, n.comp=30)
sylva.val.ica$X%*%sylva.ica$K%*%sylva.ica$W -> sylva.val.dataica
sylva.valid.label = read.table("sylva_valid.labels")
sylva.label = read.table("sylva_train.labels")
write.matrix.csr(as.matrix.csr(sylva.val.dataica), y=as.matrix(sylva.valid.label), file="sylva.ica30.valid")
write.matrix.csr(as.matrix.csr(sylva.ica$S), y=as.matrix(sylva.label), file="sylva.ica30.train")

nova <- read.table("nova_train.new")
nova.val <- read.table("nova_valid.new")

# .new檔將很少出現的維度刪去(ex.出現次數少於五次，捨掉此維度)，
# 把它減為四千多維

nova.valm = as.matrix(nova.val)
novam = as.matrix(nova)
nova.ica <- fastICA(novam, n.comp=30)
nova.val.ica <- fastICA(nova.valm, n.comp=30)
nova.val.ica$X%*%nova.ica$K%*%nova.ica$W -> nova.val.dataica
nova.valid.label = read.table("nova_valid.labels")
nova.label = read.table("nova_train.labels")
write.matrix.csr(as.matrix.csr(nova.val.dataica), y=as.matrix(nova.valid.label), file="nova.ica30.valid")
write.matrix.csr(as.matrix.csr(nova.ica$S), y=as.matrix(nova.label),	file="nova.ica30.train")
	
===========fig. 1 產生方式：==============
x1 <- rnorm(5000,mean=1,sd=1.2)
x2 <- rnorm(5000,mean=10,sd=1.2)
x3 <- rnorm(5000,mean=5)

plot(density(x1),xlim=c(-2,12),ylim=c(0,0.4),col="RED",xlab="",ylab="",main="",lwd=3)
lines(density(x2),col="RED",lwd=3)
lines(density(x3),col="Blue",lwd=3)

	
==========================================
e1071
fastICA

library(randomForest)
model <- randomForest(train_x, as.factor(train_y))
predict(model, ur_data)
write(predict(model, valid),"ada_valid.resu")