-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathFinal_Project_real_data.Rmd
292 lines (243 loc) · 13.2 KB
/
Final_Project_real_data.Rmd
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
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
---
title: "Final Project Real Dataset Part"
output: word_document
---
```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```
```{r}
#Read the training and testing data set
ziptrain<-read.table(file = "C:/Users/yinru/Dropbox/UM Biostatistics/ECE730 Statistical Learning/Final Project/zip.train.txt")
ziptest<-read.table(file = "C:/Users/yinru/Dropbox/UM Biostatistics/ECE730 Statistical Learning/Final Project/zip.test.txt")
```
#Real Data: without feature selection
```{r}
#10-fold cv
set.seed(25592732)
cv.folds <- function (n, folds = 10) {split(sample(1:n), rep(1:folds, length = n)) }
cvfolds<-cv.folds(n=dim(ziptrain)[[1]])
#Try our classifier without feature selection for real data
#K=10, p=256
#Using package "rpart"
#install.packages("rpart")
#The only porameter in tree is depth/tree size min=1,max=30
library(rpart)
#Calculate missclassification error for each model (different maxdepth)
tree.zip.err<-matrix(NA,10,1)
tree.zip.err.mean<-matrix(NA,30,1)
#could have used lapply, much faster
for (j in 1:30) {
for (i in 1:10) {
tree.zip<-rpart(V1 ~ .,
method="class", data=ziptrain[-cvfolds[[i]],],control=rpart.control(maxdepth = j))
tree.zip.pred = predict(tree.zip, newdata=ziptrain[cvfolds[[i]],], type="class")
ktest<-as.matrix(ziptrain[,1])[cvfolds[[i]],]
tree.zip.err[i,1] <- mean((ktest != tree.zip.pred))
}
tree.zip.err.mean[j,1]<-mean(tree.zip.err[,1])
}
#Find the maxdepth with the smallest cross-validated misclassification error
depth<-which.min(tree.zip.err.mean)
paste("The parameter associated with minimum error is",depth,".")
paste("The corresponding misclassification error is",tree.zip.err.mean[depth,1],".")
#display the corresponding model
tree.zip.final<-rpart(V1 ~ .,
method="class", data=ziptrain,control=rpart.control(maxdepth = depth))
printcp(tree.zip.final)
#visualize the cross-validation results
#plotcp(tree.train1)
#show detailed summary of splits
#summary(tree.train1)
# plot the tree.train1
plot(tree.zip.final, uniform=TRUE, main="Classification Tree for Train1")
text(tree.zip.final, use.n=FALSE, all=TRUE, cex=.8)
```
#Another way
##Without feature selection
```{r}
#A simpler and much faster way:
fit.tree <- rpart(V1~.,method="class", data=ziptrain)
#png(filename="C:/Users/yinru/Dropbox/UM Biostatistics/ECE730 Statistical Learning/Final #Project/3.png")
printcp(fit.tree)
plot(fit.tree, uniform=TRUE)
text(fit.tree, use.n=FALSE, all=TRUE, cex=.5)
#plotcp(fit.tree)
pred = predict(fit.tree, newdata=ziptest, type="class")
ktest<-as.matrix(ziptest[,1])
tree.zip.err <- mean((ktest != pred))
tree.zip.err
```
#Real Data:with feature selection
```{r}
#Calculate the CoV.
#Should use mean of absolute values; no unifying standard in choosing a cut-off value.
#cut-off: CoV<1
var<-NULL
cov_zip<-matrix(NA,256,1)
for(i in 1:256) {
cov_zip[i,1]<-sd(ziptrain[,i+1])/mean(abs(ziptrain[,i+1]))
if (cov_zip[i,1]<1) {
print(paste("The selected variable is x",i,sep = ""))
}
}
var<-which(cov_zip<1)
var<-var+1
var_p<-as.character(paste("V",var,sep = ""))
var_p<-data.frame(var_p)
var_p$var_p<-as.character(var_p$var_p)
#Generate regression section (text) in the model
regression<-NULL
for(i in 1:dim(var_p)[1]){
regression<-paste(regression,var_p[i,1],sep = "+")
}
regression
```
```{r}
#10-fold cv
set.seed(25592732)
cv.folds <- function (n, folds = 10) {split(sample(1:n), rep(1:folds, length = n)) }
cvfolds2<-cv.folds(n=dim(ziptrain)[[1]])
#Try our classifier with feature selection for train 1
#x1 is selected
#K=3, p=3, p1=2
#Using package "rpart"
#install.packages("rpart")
#The only porameter in tree is depth/tree size min=1,max=30
library(rpart)
#Calculate missclassification error for each model (different maxdepth)
tree.zip.err.fs2<-matrix(NA,10,1)
tree.zip.err.mean.fs2<-matrix(NA,30,1)
for (j in 1:30) {
for (i in 1:10) {
tree.zip.fs2 <- rpart(V1 ~ V2+V3+V4+V5+V6+V7+V8+V12+V13+V14+V15+V16+V17+V18+V19+V20+V21+V22+V25+V26+V29+V30+V31+V32+V33+V34+V35+V36+V37+V46+V47+V48+V49+V50+V51+V52+V53+V62+V63+V64+V65+V66+V67+V68+V69+V72+V73+V74+V75+V78+V79+V80+V81+V82+V83+V84+V85+V88+V89+V90+V94+V95+V96+V97+V98+V99+V100+V101+V102+V103+V104+V109+V110+V111+V112+V113+V114+V115+V116+V117+V118+V119+V120+V125+V126+V127+V128+V129+V130+V131+V132+V133+V134+V135+V136+V141+V142+V143+V144+V145+V146+V147+V148+V149+V150+V151+V152+V153+V157+V158+V159+V160+V161+V162+V163+V164+V165+V166+V167+V168+V169+V172+V173+V174+V175+V176+V177+V178+V179+V180+V181+V182+V183+V184+V188+V189+V190+V191+V192+V193+V194+V195+V196+V197+V198+V199+V200+V206+V207+V208+V209+V210+V211+V212+V213+V222+V223+V224+V225+V226+V227+V228+V229+V230+V233+V234+V237+V238+V239+V240+V241+V242+V243+V244+V245+V246+V247+V252+V253+V254+V255+V256+V257, method="class",data=ziptrain[-cvfolds2[[i]],], control=rpart.control(maxdepth = j))
tree.zip.pred.fs2 = predict(tree.zip.fs2, newdata=ziptrain[cvfolds2[[i]],], type="class")
ktest.fs2<-as.matrix(ziptrain[,1])[cvfolds2[[i]],]
tree.zip.err.fs2[i,1] <- mean((ktest.fs2 != tree.zip.pred.fs2))
}
tree.zip.err.mean.fs2[j,1]<-mean(tree.zip.err.fs2[,1])
}
#Find the maxdepth with the smallest cross-validated misclassification error
depth<-which.min(tree.zip.err.mean.fs2)
paste("The parameter associated with minimum error is",depth,".")
paste("The corresponding misclassification error is",tree.zip.err.mean.fs2[depth,1],".")
#display the corresponding final model
tree.zip.final.fs2<-rpart(V1 ~V2+V3+V4+V5+V6+V7+V8+V12+V13+V14+V15+V16+V17+V18+V19+V20+V21+V22+V25+V26+V29+V30+V31+V32+V33+V34+V35+V36+V37+V46+V47+V48+V49+V50+V51+V52+V53+V62+V63+V64+V65+V66+V67+V68+V69+V72+V73+V74+V75+V78+V79+V80+V81+V82+V83+V84+V85+V88+V89+V90+V94+V95+V96+V97+V98+V99+V100+V101+V102+V103+V104+V109+V110+V111+V112+V113+V114+V115+V116+V117+V118+V119+V120+V125+V126+V127+V128+V129+V130+V131+V132+V133+V134+V135+V136+V141+V142+V143+V144+V145+V146+V147+V148+V149+V150+V151+V152+V153+V157+V158+V159+V160+V161+V162+V163+V164+V165+V166+V167+V168+V169+V172+V173+V174+V175+V176+V177+V178+V179+V180+V181+V182+V183+V184+V188+V189+V190+V191+V192+V193+V194+V195+V196+V197+V198+V199+V200+V206+V207+V208+V209+V210+V211+V212+V213+V222+V223+V224+V225+V226+V227+V228+V229+V230+V233+V234+V237+V238+V239+V240+V241+V242+V243+V244+V245+V246+V247+V252+V253+V254+V255+V256+V257,
method="class", data=ziptrain, control=rpart.control(maxdepth = depth))
printcp(tree.zip.final.fs2)
# plot the tree.train1
plot(tree.zip.final.fs2, uniform=TRUE, main="Classification Tree for Train1")
text(tree.zip.final.fs2, use.n=TRUE, all=TRUE, cex=.8)
```
#Prediction error on test data
#Both models w. and w.o feature selection
```{r}
tree.zip.pred.test1 = predict(tree.zip.final, newdata=ziptest, type="class")
ktest.1<-as.matrix(ziptest[,1])
tree.zip.err.final.1 <- mean(ktest.1 != tree.zip.pred.test1)
paste("The prediction error for our classifier without feature selection is", tree.zip.err.final.1)
tree.zip.pred.test2 = predict(tree.zip.final.fs2, newdata=ziptest, type="class")
ktest.2<-as.matrix(ziptest[,1])
tree.zip.err.final.2 <- mean(ktest.2 != tree.zip.pred.test2)
paste("The prediction error for our classifier with feature selection is", tree.zip.err.final.2)
```
#Real Data:with feature selection
#A different feature selection method:cov_zip>0.4
```{r}
#Calculate the CoV.
#Should use mean of absolute values; no unifying standard in choosing a cut-off value.
#cut-off: CoV<1
var<-NULL
cov_zip<-matrix(NA,256,1)
for(i in 1:256) {
cov_zip[i,1]<-sd(ziptrain[,i+1])/mean(abs(ziptrain[,i+1]))
#if (cov_zip[i,1]<1) {
#print(paste("The selected variable is x",i,sep = ""))
#}
}
var<-which(cov_zip>0.4 & cov_zip<1)
var<-var+1
var_p<-as.character(paste("V",var,sep = ""))
var_p<-data.frame(var_p)
var_p$var_p<-as.character(var_p$var_p)
#Generate regression section (text) in the model
regression2<-NULL
for(i in 1:dim(var_p)[1]){
regression2<-paste(regression2,var_p[i,1],sep = "+")
}
regression2
```
#>0.4
```{r}
fit.tree2 <- rpart(V1 ~ V6+V7+V8+V12+V13+V14+V20+V21+V22+V25+V26+V29+V30+V31+V36+V37+V46+V47+V52+V53+V62+V63+V67+V68+V69+V72+V73+V74+V75+V78+V79+V83+V84+V85+V88+V89+V90+V94+V95+V96+V99+V100+V101+V102+V103+V104+V109+V110+V111+V112+V115+V116+V117+V118+V119+V120+V125+V126+V127+V128+V131+V132+V133+V134+V135+V136+V141+V142+V143+V144+V147+V148+V149+V150+V151+V152+V153+V157+V158+V159+V160+V163+V164+V165+V166+V167+V168+V169+V172+V173+V174+V175+V176+V179+V180+V181+V182+V183+V184+V188+V189+V190+V191+V192+V195+V196+V197+V198+V199+V200+V206+V207+V208+V211+V212+V213+V222+V223+V224+V228+V229+V230+V233+V234+V237+V238+V239+V246+V247+V252+V253,
method="class", data=ziptrain)
#png(filename="C:/Users/yinru/Dropbox/UM Biostatistics/ECE730 Statistical Learning/Final #Project/4.png")
printcp(fit.tree2)
plot(fit.tree2, uniform=TRUE)
text(fit.tree2, use.n=FALSE, all=TRUE, cex=.5)
#plotcp(fit.tree2)
#rsq.rpart(fit.tree1)
pred = predict(fit.tree2, newdata=ziptest, type="class")
ktest<-as.matrix(ziptest[,1])
tree.zip.err <- mean((ktest != pred))
tree.zip.err
```
#>0.05
```{r}
fit.tree2 <- rpart(V1 ~ V2+V3+V4+V5+V6+V7+V8+V12+V13+V14+V15+V16+V17+V18+V19+V20+V21+V22+V25+V26+V29+V30+V31+V32+V33+V34+V35+V36+V37+V46+V47+V48+V49+V50+V51+V52+V53+V62+V63+V64+V65+V66+V67+V68+V69+V72+V73+V74+V75+V78+V79+V80+V81+V82+V83+V84+V85+V88+V89+V90+V94+V95+V96+V97+V98+V99+V100+V101+V102+V103+V104+V109+V110+V111+V112+V113+V114+V115+V116+V117+V118+V119+V120+V125+V126+V127+V128+V129+V130+V131+V132+V133+V134+V135+V136+V141+V142+V143+V144+V145+V146+V147+V148+V149+V150+V151+V152+V153+V157+V158+V159+V160+V161+V162+V163+V164+V165+V166+V167+V168+V169+V172+V173+V174+V175+V176+V177+V178+V179+V180+V181+V182+V183+V184+V188+V189+V190+V191+V192+V193+V194+V195+V196+V197+V198+V199+V200+V206+V207+V208+V209+V210+V211+V212+V213+V222+V223+V224+V225+V226+V227+V228+V229+V230+V233+V234+V237+V238+V239+V240+V241+V243+V244+V245+V246+V247+V252+V253+V254+V255+V256+V257,
method="class", data=ziptrain
)
printcp(fit.tree2)
plot(fit.tree2, uniform=TRUE)
text(fit.tree2, use.n=FALSE, all=TRUE, cex=.5)
plotcp(fit.tree2)
#rsq.rpart(fit.tree1)
pred = predict(fit.tre21, newdata=ziptest, type="class")
ktest<-as.matrix(ziptest[,1])
tree.zip.err <- mean((ktest != pred))
tree.zip.err
```
#>0.15
```{r}
fit.tree3 <- rpart(V1 ~ V3+V4+V5+V6+V7+V8+V12+V13+V14+V15+V16+V19+V20+V21+V22+V25+V26+V29+V30+V31+V32+V33+V35+V36+V37+V46+V47+V48+V49+V50+V51+V52+V53+V62+V63+V64+V65+V66+V67+V68+V69+V72+V73+V74+V75+V78+V79+V80+V82+V83+V84+V85+V88+V89+V90+V94+V95+V96+V97+V98+V99+V100+V101+V102+V103+V104+V109+V110+V111+V112+V113+V114+V115+V116+V117+V118+V119+V120+V125+V126+V127+V128+V129+V130+V131+V132+V133+V134+V135+V136+V141+V142+V143+V144+V145+V146+V147+V148+V149+V150+V151+V152+V153+V157+V158+V159+V160+V161+V162+V163+V164+V165+V166+V167+V168+V169+V172+V173+V174+V175+V176+V177+V178+V179+V180+V181+V182+V183+V184+V188+V189+V190+V191+V192+V193+V194+V195+V196+V197+V198+V199+V200+V206+V207+V208+V209+V210+V211+V212+V213+V222+V223+V224+V225+V227+V228+V229+V230+V233+V234+V237+V238+V239+V240+V241+V244+V245+V246+V247+V252+V253+V254+V255+V256,
method="class", data=ziptrain
)
printcp(fit.tree3)
plot(fit.tree3, uniform=TRUE)
text(fit.tree3, use.n=FALSE, all=TRUE, cex=.5)
plotcp(fit.tree3)
#rsq.rpart(fit.tree1)
pred3 = predict(fit.tree3, newdata=ziptest, type="class")
ktest<-as.matrix(ziptest[,1])
tree.zip.err3 <- mean((ktest != pred3))
tree.zip.err3
```
#>0.25
```{r}
fit.tree4 <- rpart(V1 ~ V4+V5+V6+V7+V8+V12+V13+V14+V15+V19+V20+V21+V22+V25+V26+V29+V30+V31+V32+V35+V36+V37+V46+V47+V48+V51+V52+V53+V62+V63+V64+V67+V68+V69+V72+V73+V74+V75+V78+V79+V80+V83+V84+V85+V88+V89+V90+V94+V95+V96+V99+V100+V101+V102+V103+V104+V109+V110+V111+V112+V114+V115+V116+V117+V118+V119+V120+V125+V126+V127+V128+V129+V130+V131+V132+V133+V134+V135+V136+V141+V142+V143+V144+V145+V146+V147+V148+V149+V150+V151+V152+V153+V157+V158+V159+V160+V161+V162+V163+V164+V165+V166+V167+V168+V169+V172+V173+V174+V175+V176+V177+V178+V179+V180+V181+V182+V183+V184+V188+V189+V190+V191+V192+V193+V194+V195+V196+V197+V198+V199+V200+V206+V207+V208+V211+V212+V213+V222+V223+V224+V227+V228+V229+V230+V233+V234+V237+V238+V239+V240+V245+V246+V247+V252+V253+V254,
method="class", data=ziptrain
)
printcp(fit.tree4)
plot(fit.tree4, uniform=TRUE)
text(fit.tree4, use.n=FALSE, all=TRUE, cex=.5)
plotcp(fit.tree4)
#rsq.rpart(fit.tree1)
pred4 = predict(fit.tree4, newdata=ziptest, type="class")
ktest<-as.matrix(ziptest[,1])
tree.zip.err4 <- mean((ktest != pred4))
tree.zip.err4
```
#>0.9
```{r}
fit.tree6 <- rpart(V1 ~ V7+V8+V12+V22+V25+V26+V29+V30+V37+V46+V53+V62+V69+V72+V73+V74+V75+V78+V84+V85+V88+V89+V90+V94+V100+V101+V102+V103+V104+V109+V110+V116+V117+V118+V119+V120+V125+V126+V127+V132+V133+V134+V135+V136+V141+V142+V143+V148+V149+V150+V151+V152+V153+V157+V158+V159+V164+V165+V166+V167+V168+V169+V172+V173+V174+V175+V180+V181+V182+V183+V184+V188+V189+V190+V191+V196+V197+V198+V199+V200+V206+V207+V212+V213+V222+V229+V230+V233+V234+V237+V247+V252,
method="class", data=ziptrain
)
printcp(fit.tree6)
plot(fit.tree6, uniform=TRUE)
text(fit.tree6, use.n=FALSE, all=TRUE, cex=.5)
plotcp(fit.tree6)
#rsq.rpart(fit.tree1)
pred6 = predict(fit.tree6, newdata=ziptest, type="class")
ktest<-as.matrix(ziptest[,1])
tree.zip.err6 <- mean((ktest != pred6))
tree.zip.err6
```