-
Notifications
You must be signed in to change notification settings - Fork 0
/
Question 04-cao.R
136 lines (127 loc) · 4.42 KB
/
Question 04-cao.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
# Question 04
library(readr)
library(dplyr)
library(magrittr)
# Random Forest -----------------------------------------------------------
control<-rpart.control(minbucket=3,minsplit=10,xval=114)
#讀取資料
data <- read_csv("Glass_data.csv")
data %<>% as.data.frame()
data[,10] %<>% as.factor()
num <- sample(1:114,size = 92,replace = F)
traindata <- data[num,]
testdata <- data[-num,]
table(data$Type)
# 載入隨機樹森林package
library(randomForest)
randomForest(x = data[,-10], y = data$Type, ntree = 500, proximity = T, data = data, importane = T)
# 決定多少個節點
tuneRF(data[,-10], data[,10])
# 跑隨機樹森林模型(節點數=根號(變數個數))
model <- rfcv(trainx = data[,-10],
trainy = data$Type,
data = data,
importane = T,
proximity = T,
nodesize = 3,
step = 0.9,
ntree = 500,
cv.fold = 5)
model$importance
# 跑隨機樹森林模型(變數改為9個)
model <- randomForest(formula = Type ~ .,
data = traindata,
mtry = 9,
nodesize=1,
importane = T,
proximity = T,
ntree = 500,cv.fold = 114)
model
model$votes
model$predicted
# leave-one-out
lou <- rfcv(traindata[,-10], traindata[,10], step=0.7)
lou$error.cv
which(lou$error.cv == min(lou$error.cv))
# In this case, the minimum true error rate is about 0.30 when 6 predictors are used to split the node
# 混淆
C <- table(model$predicted,traindata$Type) ; C
# apparent error rate (測試樣本的錯誤率)
1-sum(diag(C))/nrow(traindata)
# 森林中每棵樹的節點個數
hist(treesize(model, terminal=TRUE))
# 錯誤率: 利用OOB(Out Of Bag)運算出來的
plot(model)
# 衡量每一個變數對Y值的重要性，取到小數點第二位
round(importance(model), 2)
# 分對分錯
library(cluster)
clusters <- pam(1-model$proximity, k=6, diss = TRUE)
table(clusters$clustering,traindata$Type)
# 畫圖
# 群集分析(投影在二維)
mds <- MDSplot(rf = model,
fac = data$Type,
k = 2,
pch = 1)
mds$GOF
mds$points
# Boosting ----------------------------------------------------------------
#install.packages("adabag")
library(adabag)
#glass.control<-rpart.control(minbucket=3,minsplit=10,xval=100)
traindata.boost <- boosting(Type~.,
data=traindata,
boos = T,
mfinal=10 )
traindata.boost$importance
test <- predict.boosting(traindata.boost,traindata)
test$confusion
traindata.pred <- predict.boosting(traindata.boost,testdata)
# 混淆矩陣
traindata.pred$confusion
# 真實錯誤率
C <- traindata.pred$confusion
1-sum(diag(C))/nrow(testdata)
# Boosting.cv -------------------------------------------------------------
boostingmodel <- boosting.cv(Type~.,
data = data,
v = 114,
boos = T,
mfinal = 10)
boostingmodel
# xgboost -----------------------------------------------------------------
# xgboost
library(useful)
# 形容模型的公式
# 因為是建立樹模型，我們不需要截距項
Formula <- Type~refractive+sodium+Magnesium+Aluminum+Silicon+Potassium+Calcium+Barium+Iron-1
# 因為是建立樹模型，我們使用所有level的類別變數
glassX <- build.x(Formula,data = data,contrasts = F)
glassY <- build.y(Formula,data = data)
num_class <- 7
library(xgboost)
# xgboost 參數設定 (xgboost parameters setup)
param = list("objective" = "multi:softmax", # 訓練目標
"eval_metric" = "merror", # 多分類錯分率 (多分類對數損失用mlogloss)
"num_class" = num_class
)
glassboost <- xgboost(data = glassX,
label = glassY,
nrounds = 20,
params = param,
nfold = 114,
boos = T,
control=)
glassboost
xgb.plot.multi.trees(glassboost,feature_names = colnames(glassX))
# 重要變數
xgb.plot.importance(xgb.importance(glassboost,
feature_names = colnames(glassX)))
xdata <- build.x(Formula,data = testdata,contrasts = FALSE)
# 計算預測值 (get prediction)
Ypred <- predict(glassboost,xdata)
# 混淆矩陣 (confusion matrix)
C <- table(testdata$Type,Ypred);C
# 真實錯誤率
1-sum(diag(C))/sum(C)