-
Notifications
You must be signed in to change notification settings - Fork 0
/
CV_SVM_rad.R
156 lines (138 loc) · 5.08 KB
/
CV_SVM_rad.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
#####################################################################
# function to train SVM with radial kernel
#####################################################################
CV_SVM_rad<-function(x,y,K,cost,gamma,nb_cores,seed) {
library(e1071)
library(foreach)
library(doMC)
registerDoMC(nb_cores)
set.seed(seed)
out<-list()
out$fit<-list()
nb_folds<-length(K)
for (i in 1:nb_folds)
eval(parse(text=paste("x_",i,"<-as.matrix(if (length(x)!=nb_folds) x else x[[i]])",sep="")))
rm(list="x")
out$fit <- foreach (k = 1:nb_folds) %dopar% {
set.seed(seed+k)
eval(parse(text=paste("svm(x=x_",k,"[-K[[k]],],y=y[-K[[k]]],
cost=cost,gamma=gamma,kernel='radial',scale=FALSE)",sep="")))
}
out$yhatV<-rep(0,length(y))
for (i in 1:length(K))
eval(parse(text=paste("
out$yhatV[K[[i]]]<-predict(out$fit[[i]],x_",i,"[K[[i]],])",sep="")))
out
}
#####################################################################
# function to fine tune SVM with radial kernel
#####################################################################
tune_SVM_rad<-function(x,y,K,cost_i,gamma_i,cost_var,gamma_var,step_max,nb_cores,seed) {
set.seed(seed)
res_mat<-matrix(NA,31,31)
colnames(res_mat)<-round(gamma_i*gamma_var^(-15:15),12)
rownames(res_mat)<-round(cost_i*cost_var^(-15:15),12)
plot(gamma_i,cost_i,pch=19,log="xy",col=4,
xlab="gamma",ylab="cost",
xlim=range(as.numeric(colnames(res_mat))),
ylim=range(as.numeric(rownames(res_mat))))
sol<-list()
sol$Improved<-"NO"
sol$pos<-NA
sol$svm_obj<-CV_SVM_rad(x,y,K,cost_i,gamma_i,nb_cores,seed)
sol$res<-abs(cor(y,sol$svm_obj$yhatV,method="pearson"))
sol$cost<-cost_i
sol$gamma<-gamma_i
print(paste("cost:",cost_i,"/ gamma:",gamma_i,"/ metric:",sol$res))
res_mat[as.character(round(cost_i,12)),as.character(round(gamma_i,12))]<-sol$res
mult<-list()
mult[[1]]<-c(gamma=gamma_var^2,cost=cost_var^2)
mult[[2]]<-c(gamma=gamma_var^2,cost=1)
mult[[3]]<-c(gamma=gamma_var^2,cost=1/cost_var^2)
mult[[4]]<-c(gamma=1,cost=1/cost_var^2)
mult[[5]]<-c(gamma=1/gamma_var^2,cost=1/cost_var^2)
mult[[6]]<-c(gamma=1/gamma_var^2,cost=1)
mult[[7]]<-c(gamma=1/gamma_var^2,cost=cost_var^2)
mult[[8]]<-c(gamma=1,cost=cost_var^2)
for (pos in 1:8) {
cost<-cost_i*mult[[pos]]['cost']
gamma<-gamma_i*mult[[pos]]['gamma']
points(gamma,cost,pch=1,col=1)
svm_obj<-CV_SVM_rad(x,y,K,cost,gamma,nb_cores,seed)
res<-abs(cor(y,svm_obj$yhatV,method="pearson"))
res_mat[as.character(round(cost,12)),as.character(round(gamma,12))]<-res
print(paste("cost:",cost,"/ gamma:",gamma,"/ metric:",res))
if (res>sol$res) {
sol$Improved<-"YES"
sol$pos<-pos
sol$res<-res
sol$svm_obj<-svm_obj
sol$cost<-cost
sol$gamma<-gamma
print(paste("Improved: YES"))
}
}
points(sol$gamma,sol$cost,pch=19,col=3)
mult[[1]]<-c(gamma=gamma_var,cost=cost_var)
mult[[2]]<-c(gamma=gamma_var,cost=1)
mult[[3]]<-c(gamma=gamma_var,cost=1/cost_var)
mult[[4]]<-c(gamma=1,cost=1/cost_var)
mult[[5]]<-c(gamma=1/gamma_var,cost=1/cost_var)
mult[[6]]<-c(gamma=1/gamma_var,cost=1)
mult[[7]]<-c(gamma=1/gamma_var,cost=cost_var)
mult[[8]]<-c(gamma=1,cost=cost_var)
if (sol$Improved=="NO") {
i<-sample(1:8,1)
for (k in 1:8) {
pos<-1+(i+(k)*5)%%8
cost<-cost_i*mult[[pos]]['cost']
gamma<-gamma_i*mult[[pos]]['gamma']
points(gamma,cost,pch=1,col=1)
svm_obj<-CV_SVM_rad(x,y,K,cost,gamma,nb_cores,seed)
res<-abs(cor(y,svm_obj$yhatV,method="pearson"))
res_mat[as.character(round(cost,12)),as.character(round(gamma,12))]<-res
print(paste("cost:",cost,"/ gamma:",gamma,"/ metric:",res))
if (res>sol$res) {
sol$Improved<-"YES"
sol$pos<-pos
sol$res<-res
sol$svm_obj<-svm_obj
sol$cost<-cost
sol$gamma<-gamma
points(gamma,cost,pch=19,col=3)
print(paste("Improved: YES"))
break
}
}
}
if (sol$Improved=="YES") for (i in 1:step_max) {
sol$Improved<-"NO"
pos<-sol$pos
for (k in 1:8) {
pos<-1+(-1+pos+(k-1)*(-1)^k)%%8
cost<-sol$cost*mult[[pos]]['cost']
gamma<-sol$gamma*mult[[pos]]['gamma']
points(gamma,cost,pch=1,col=1)
if (is.na(res_mat[as.character(round(cost,12)),as.character(round(gamma,12))])) {
svm_obj<-CV_SVM_rad(x,y,K,cost,gamma,nb_cores,seed)
res<-abs(cor(y,svm_obj$yhatV,method="pearson"))
res_mat[as.character(round(cost,12)),as.character(round(gamma,12))]<-res
print(paste("cost:",cost,"/ gamma:",gamma,"/ metric:",res))
if (res>sol$res) {
sol$Improved<-"YES"
sol$pos<-pos
sol$res<-res
sol$svm_obj<-svm_obj
sol$cost<-cost
sol$gamma<-gamma
points(gamma,cost,pch=19,col=3)
print(paste("Improved: YES"))
break
}
}
}
}
print(paste("best cost",sol$cost,"best gamma",sol$gamma,"/ metric:",sol$res))
points(sol$gamma,sol$cost,pch=19,cex=2,col=2)
sol$svm_obj
}