-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathnnEnsembleOutliers.R
235 lines (200 loc) · 6.64 KB
/
nnEnsembleOutliers.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
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
#install.packages("rminer")
library(party)
library(rminer)
library(adabag)
setwd('C:\\Users\\BlueMoon\\Documents\\GitHub\\kdd')
d=read.csv('training.csv', TRUE, ';')
test=read.csv('test.csv', TRUE, ';')
# Add new features
d$X31 <- d$X1 * d$X22
d$X32 <- d$X1 * d$X28
d$X33 <- d$X3 * d$X10
d$X34 <- d$X3 * d$X18
d$X35 <- d$X3 * d$X22
d$X36 <- d$X3 * d$X28
d$X37 <- d$X6 * d$X7
d$X38 <- d$X6 * d$X8
d$X39 <- d$X6 * d$X18
d$X40 <- d$X7 * d$X12
d$X41 <- d$X7 * d$X19
d$X42 <- d$X7 * d$X24
d$X43 <- d$X7 * d$X30
d$X44 <- d$X8 * d$X12
d$X45 <- d$X8 * d$X19
d$X46 <- d$X8 * d$X23
d$X47 <- d$X8 * d$X24
d$X48 <- d$X8 * d$X30
d$X49 <- d$X12 * d$X17
d$X50 <- d$X12 * d$X18
d$X51 <- d$X13 * d$X21
d$X52 <- d$X17 * d$X19
d$X53 <- d$X18 * d$X19
d$X54 <- d$X18 * d$X30
d$X55 <- d$X21 * d$X23
d$X56 <- d$X23 * d$X25
d$X57 <- d$X23 * d$X26
test$X31 <- test$X1 * test$X22
test$X32 <- test$X1 * test$X28
test$X33 <- test$X3 * test$X10
test$X34 <- test$X3 * test$X18
test$X35 <- test$X3 * test$X22
test$X36 <- test$X3 * test$X28
test$X37 <- test$X6 * test$X7
test$X38 <- test$X6 * test$X8
test$X39 <- test$X6 * test$X18
test$X40 <- test$X7 * test$X12
test$X41 <- test$X7 * test$X19
test$X42 <- test$X7 * test$X24
test$X43 <- test$X7 * test$X30
test$X44 <- test$X8 * test$X12
test$X45 <- test$X8 * test$X19
test$X46 <- test$X8 * test$X23
test$X47 <- test$X8 * test$X24
test$X48 <- test$X8 * test$X30
test$X49 <- test$X12 * test$X17
test$X50 <- test$X12 * test$X18
test$X51 <- test$X13 * test$X21
test$X52 <- test$X17 * test$X19
test$X53 <- test$X18 * test$X19
test$X54 <- test$X18 * test$X30
test$X55 <- test$X21 * test$X23
test$X56 <- test$X23 * test$X25
test$X57 <- test$X23 * test$X26
test$STATUS <- as.factor(c(rep(0, 50)))
d=d[,c(2:length(d))]
#print(summary(d))
remove_outliers <- function(x, val, na.rm = TRUE, ...) {
qnt <- quantile(x, probs=c(val, 1-val), na.rm = na.rm, ...)
H <- 1.5 * IQR(x, na.rm = na.rm)
y <- x
y[x < (qnt[1] - H)] <- NA
y[x > (qnt[2] + H)] <- NA
y
}
attributes=colnames(d)[c(2:length(d))]
d.o <- d
for(i in 1:length(attributes)){
d.o[[attributes[i]]]=remove_outliers(d.o[[attributes[i]]], 0.05)
}
#remove outliers
d.n <- d[complete.cases(d.o), ]
d.r <- d[!complete.cases(d.o), ]
test.o <- test
for(i in 1:length(attributes)){
test.o[[attributes[i]]]=remove_outliers(test.o[[attributes[i]]], 0.05)
}
#remove outliers
test.n <- test[complete.cases(test.o), ]
test.r <- test[!complete.cases(test.o), ]
# MLP
NNmine.n=mining(STATUS~.,d.n,model="mlpe",Runs=5,method=c("kfold",3),search="heuristic5",feat="s")
NN.n=fit(STATUS~.,d.n,model="mlpe",search=NNmine.n$mpar)
PNN.n=predict(NN.n,test.n)
NNmine.r=mining(STATUS~.,d.r,model="mlpe",Runs=5,method=c("kfold",3),search="heuristic5",feat="s")
NN.r=fit(STATUS~.,d.r,model="mlpe",search=NNmine.r$mpar)
PNN.r=predict(NN.r,test.r)
# SVM
SVmine.n=mining(STATUS~.,d.n,model="ksvm",Runs=5,method=c("kfold",3),search="heuristic5",f="s")
SV.n=fit(STATUS~.,d.n,model="ksvm",search=SVmine.n$mpar) # fit the SVM
PSVM.n=predict(SV.n,test.n)
SVmine.r=mining(STATUS~.,d.r,model="ksvm",Runs=5,method=c("kfold",3),search="heuristic5",f="s")
SV.r=fit(STATUS~.,d.r,model="ksvm",search=SVmine.r$mpar) # fit the SVM
PSVM.r=predict(SV.r,test.r)
# CIF
CIF.n <- cforest(STATUS~., data = d.n, controls=cforest_unbiased(ntree=100, mtry=3))
PCIF.n <- predict(CIF.n, test.n, OOB=TRUE, type = "response")
CIF.r <- cforest(STATUS~., data = d.r, controls=cforest_unbiased(ntree=100, mtry=3))
PCIF.r <- predict(CIF.r, test.r, OOB=TRUE, type = "response")
# RF
d.n$STATUS <- as.factor(d.n$STATUS)
RF.n=fit(STATUS~.,d.n,model="randomforest") # fit a random forest
PRFu.n=predict(RF.n,test.n)
PRF.n=c(1:50)
for(i in 1:length(PRFu.n[,1])) {
PRF.n[i] <- ifelse(PRFu.n[i,1] < PRFu.n[i,2],PRFu.n[i,2]-PRFu.n[i,1],-(PRFu.n[i,1]-PRFu.n[i,2]))
}
d.r$STATUS <- as.factor(d.r$STATUS)
RF.r=fit(STATUS~.,d.r,model="randomforest") # fit a random forest
PRFu.r=predict(RF.r,test.r)
PRF.r=c(1:50)
for(i in 1:length(PRFu.r[,1])) {
PRF.r[i] <- ifelse(PRFu.r[i,1] < PRFu.r[i,2],PRFu.r[i,2]-PRFu.r[i,1],-(PRFu.r[i,1]-PRFu.r[i,2]))
}
# kNN
KNNmine.n=mining(STATUS~.,d.n,model="kknn",Runs=5,method=c("kfold",3),search="heuristic5",f="s")
KNN.n=fit(STATUS~.,d.n,model="kknn",search=KNNmine.n$mpar)
PKNN.n=predict(KNN.n,test.n)
KNNmine.r=mining(STATUS~.,d.r,model="kknn",Runs=5,method=c("kfold",3),search="heuristic5",f="s")
KNN.r=fit(STATUS~.,d.r,model="kknn",search=KNNmine.r$mpar)
PKNN.r=predict(KNN.r,test.r)
# Boosting
#d$STATUS <- as.factor(d$STATUS)
test.n$STATUS <- as.factor(c(rep(1, length(test.n[[1]])/2),rep(-1,length(test.n[[1]])/2)))
BO.n <- boosting(STATUS ~ ., data = d.n, coeflearn="Zhu")
PBOu.n <- (predict(BO.n, newdata=test.n, type="class"))$prob
PBO.n=c(1:length(PBOu.n[,1]))
for(i in 1:length(PBOu.n[,1])) {
PBO.n[i] <- ifelse(PBOu.n[i,1] < PBOu.n[i,2],PBOu.n[i,2]-PBOu.n[i,1],-(PBOu.n[i,1]-PBOu.n[i,2]))
}
test.r$STATUS <- as.factor(c(rep(1, length(test.r[[1]])/2),rep(-1,length(test.r[[1]])/2)))
BO.r <- boosting(STATUS ~ ., data = d.r, coeflearn="Zhu")
PBOu.r <- (predict(BO.r, newdata=test.r, type="class"))$prob
PBO.r=c(1:length(PBOu.r[,1]))
for(i in 1:length(PBOu.r[,1])) {
PBO.r[i] <- ifelse(PBOu.r[i,1] < PBOu.r[i,2],PBOu.r[i,2]-PBOu.r[i,1],-(PBOu.r[i,1]-PBOu.r[i,2]))
}
PRF=c(1:50)
PNN=c(1:50)
PSVM=c(1:50)
PBO=c(1:50)
PKNN=c(1:50)
PCIF=c(1:50)
counter = 1
i=1
j=1
while(counter<=50){
if((i<=length(test.n$ID) || j>length(test.r$ID)) && test.n$ID[i] < test.r$ID[j]){
PRF[counter]=PRF.n[i]
PNN[counter]=PNN.n[i]
PSVM[counter]=PSVM.n[i]
PBO[counter]=PBO.n[i]
PKNN[counter]=PKNN.n[i]
PCIF[counter]=PCIF.n[i]
i=i+1
}else{
PRF[counter]=PRF.r[j]
PNN[counter]=PNN.r[j]
PSVM[counter]=PSVM.r[j]
PBO[counter]=PBO.r[j]
PKNN[counter]=PKNN.r[j]
PCIF[counter]=PCIF.r[j]
j=j+1
}
counter=counter+1
}
# get the predictions:
results=c(1:50)
for(i in 1:length(test$STATUS)){
PRF[i] <- ifelse(PRF[i] > 0,PRF[i]^2,-(PRF[i]^2))
PNN[i] <- ifelse(PNN[i] > 0,PNN[i]^2,-(PNN[i]^2))
PSVM[i] <- ifelse(PSVM[i] > 0,PSVM[i]^2,-(PSVM[i]^2))
PBO[i] <- ifelse(PBO[i] > 0,PBO[i]^2,-(PBO[i]^2))
PKNN[i] <- ifelse(PKNN[i] > 0,PKNN[i]^2,-(PKNN[i]^2))
PCIF[i] <- ifelse(PCIF[i] > 0,PCIF[i]^2,-(PCIF[i]^2))
results[i]=PRF[i]+PNN[i]+PSVM[i]+PBO[i]+PKNN[i]+PCIF[i]
}
P=data.frame(ID=test$ID,STATUS=results)
for(i in 1:length(P$STATUS)) {
P$STATUS[i] <- ifelse(P$STATUS[i] > 0,1,-1)
}
write.csv(P,"results.csv", row.names=FALSE) # save output and predictions
#REMOVE OUTLIERS
#mpause("Show average MAE metric:")
#eNN=mmetric(NN,metric="MAE")
#mi=meanint(eNN)
#cat("MAE average=:",mi$mean,"+-",mi$int,"\n")
#mpause("Show scatter plot:")
#mgraph(NN,graph="RSC",main="MLP",baseline=TRUE,Grid=TRUE)
#mpause("Show importance graph:")
#nw=c(1:31)
#mgraph(NNmine,graph="IMP",leg=nw,xval=0.0,Grid=TRUE)