-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathheart2.r
156 lines (144 loc) · 6.52 KB
/
heart2.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
# Experiments on heart disease data
library(e1071)
library(glmnet)
library(parcor)
library(msgps)
scan_heart_data <-function(filename1, nl = 0){
data1 = scan(filename1, nlines = nl, what = list(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, ""))
l = length(data1[[1]])
data1m = matrix(unlist(data1), l, 76)
matrix(as.numeric(data1m[,1:75]), l, 75)
}
#### load heart disease data (899 samples)
h1 = scan_heart_data("http://archive.ics.uci.edu/ml/machine-learning-databases/heart-disease/cleveland.data", 282*10)
h2 = scan_heart_data("http://archive.ics.uci.edu/ml/machine-learning-databases/heart-disease/hungarian.data")
h3 = scan_heart_data("http://archive.ics.uci.edu/ml/machine-learning-databases/heart-disease/switzerland.data")
h4 = scan_heart_data("http://archive.ics.uci.edu/ml/machine-learning-databases/heart-disease/long-beach-va.data")
heart1 = as.matrix( rbind(h1,h2,h3,h4) )
#### prepare data
#58 (num) (the predicted attribute)
cls = heart1[,58]
#### 13 attributes recommended by dataset creator
#3 (age) #4 (sex) #9 (cp) #10 (trestbps) #12 (chol)
#16 (fbs) #19 (restecg) #32 (thalach) #38 (exang)
#40 (oldpeak) #41 (slope) #44 (ca) #51 (thal)
data1 = heart1[,c(3,4,9,10,12,16,19,32,38,40,41,44,51)]
#### attributes selected with Copula entropy, distance correlation, dHSIC
data2 = heart1[,c(3,4,6,7,9,12,16,28:32,38,40,41,44,51,59:68)] # copula entropy
data2b = heart1[,c(3,4,6,7,9,12,13,16,28:33,38,40,41,52,59:68)] # dcor
data2c = heart1[,c(3,4,6,7,9,12,13,16,25,28:32,38,40,41,44,59:68)] # dhsic
#### for GLMs
data3 = heart1[,c(3:57,59:68)]
## training models
# SVMs
svm1 = svm(data1,cls, type = "nu-regression", gamma = 2)
svm2 = svm(data2,cls, type = "nu-regression", gamma = 2)
svm2b = svm(data2b,cls, type = "nu-regression", gamma = 2)
svm2c = svm(data2c,cls, type = "nu-regression", gamma = 2)
# GLM
lasso1 = cv.glmnet(data3, cls, family = "multinomial", type.measure = "class", nfolds = 10)
ridge1 = cv.glmnet(data3, cls, family = "multinomial", alpha = 0, type.measure = "class", nfolds = 10)
elnet1 = cv.glmnet(data3, cls, family = "multinomial", alpha = 0.5, type.measure = "class", nfolds = 10)
#### Adaptive Lasso
ada1 = adalasso(data3,cls)
ada2 = msgps(data3,cls,penalty = "alasso")
## prediction
pred1 = predict(svm1,data1)
pred2 = predict(svm2,data2)
pred2b = predict(svm2b,data2b)
pred2c = predict(svm2c,data2c)
pred3 = predict(lasso1, newx = data3, s = "lambda.min", type = "class")
pred4 = predict(ridge1, newx = data3, s = "lambda.min", type = "class")
pred5 = predict(elnet1, newx = data3, s = "lambda.min", type = "class")
# adaptive lasso
## ada1 parcor
intada1 = ada1$intercept.adalasso
coefada1 = ada1$coefficients.adalasso
predada1 = data3 %*% coefada1 + intada1
## ada2 msgps
coefada2 = coef.msgps(ada2)
predada2 = predict(ada2, data3)
## correct numbers
pred1num = sum(abs(cls-pred1)<=0.5) # recommandation
pred2num = sum(abs(cls-pred2)<=0.5) # ce
pred2bnum = sum(abs(cls-pred2b)<=0.5) # dcor
pred2cnum = sum(abs(cls-pred2c)<=0.5) # dhsic
pred3num = sum(cls==pred3) # lasso
pred4num = sum(cls==pred4) # ridge regression
pred5num = sum(cls==pred5) # elastic net
prednum1ada = sum(abs(cls-predada1)<0.5) # ada1 parcor
prednum2ada = sum(abs(cls-predada2[,3])<0.5) # ada2 msgps
#### plotting for GLMs
## coefficients of lasso
coef10 = coef(lasso1, s = "lambda.min")
coef10a = abs(coef10$'0'[2:66])+abs(coef10$'1'[2:66])+abs(coef10$'2'[2:66])+abs(coef10$'3'[2:66])+abs(coef10$'4'[2:66])
x11();
#pdf("~/Rworks/heart/lasso1.pdf")
barplot(abs(coef10a * colMeans(data3)), xlab = "Variable ID", ylab = "Coefficients Value", main = "LASSO")
axis(side = 1, at = c(1:65)*65/54-0.5, labels = c(3:57,59:68))
#dev.off()
## coefficients of ridge regression
coef40 = coef(ridge1, s = "lambda.min")
coef40a = abs(coef40$'0'[2:66])+abs(coef40$'1'[2:66])+abs(coef40$'2'[2:66])+abs(coef40$'3'[2:66])+abs(coef40$'4'[2:66])
x11();
#pdf("~/Rworks/heart/ridge1.pdf")
barplot(abs(coef40a * colMeans(data3)), xlab = "Variable ID", ylab = "Coefficients Value", main = "Ridge Regression")
axis(side = 1, at = c(1:65)*65/54-0.5, labels = c(3:57,59:68))
#dev.off()
## coefficients of elastic net
coef5 = coef(elnet1, s = "lambda.min")
coef5a = abs(coef5$'0'[2:66])+abs(coef5$'1'[2:66])+abs(coef5$'2'[2:66])+abs(coef5$'3'[2:66])+abs(coef5$'4'[2:66])
x11();
#pdf("~/Rworks/heart/elnet1.pdf")
barplot(abs(coef5a * colMeans(data3)), xlab = "Variable ID", ylab = "Coefficients Value", main = "Elastic Net")
axis(side = 1, at = c(1:65)*65/54-0.5, labels = c(3:57,59:68))
#dev.off()
## coefficients of Adaptive Lasso
# ada1 parcor
names(coefada1) = {} #c(3:57,59:68)
x11();
#pdf("~/Rworks/heart/adalasso1.pdf")
barplot(abs(coefada1), xlab = "Variable ID", ylab = "Coefficients Value", main = "Adaptive LASSO")
axis(side = 1, at = c(1:65)*65/54-0.5, labels = c(3:57,59:68))
#dev.off()
# ada2 msgps
coefada23 = abs(coefada2[2:66,3])
names(coefada23) ={}
x11(); barplot(coefada23, xlab = "Variable ID", ylab = "Coefficients Value", main = "Adaptive LASSO2")
axis(side = 1, at = c(1:65)*65/54-0.5, labels = c(3:57,59:68))
######## step_AIC(glm)
d31 = data.frame(data3)
names(d31) = c(3:57,59:68)
d3a = cbind(d31,cls)
glm1 = glm(cls~., data = d3a, family = poisson(link = "log"))
glm1s = step(glm1)
pred1glm1 = predict(glm1s, newdata = d3a, type = "response")
prednum1aic = sum(abs(cls-pred1glm1)<0.5)
cat("Accuracy of stepwise GLM (AIC) :", prednum1aic,"\n")
## plot coefficients
idx1 = c(3,4,5,9,12,16,18,26,28,29,30,32,38,40,44,47,50,53,54,60,61,63,64,65,66,67)
coef1step = rep(0,68)
coef1step[idx1] = abs(coef(glm1s)[2:27])
x11();
#pdf("~/Rworks/heart/step1.pdf")
barplot(coef1step[c(3:57,59:68)], xlab = "Variable ID", ylab = "Coefficients Value", main = "Stepwise GLM(AIC)")
axis(side = 1, at = c(1:65)*65/54-0.5, labels = c(3:57,59:68))
#dev.off()
#### step_BIC(glm)
d31 = data.frame(data3)
names(d31) = c(3:57,59:68)
d3a = cbind(d31,cls)
glm2 = glm(cls~., data = d3a, family = poisson(link = "log"))
glm2s = step(glm2, k = log(length(cls)))
pred2glm2 = predict(glm2s, newdata = d3a, type = "response")
prednum1bic = sum(abs(cls-pred2glm2)<0.5)
cat("Accuracy of stepwise GLM (BIC) :", prednum1bic,"\n")
## plot coefficients
idx2 = c(3,4,5,9,16,18,29,30,40,53,63,66,67)
coef2step = rep(0,68)
coef2step[idx2] = abs(coef(glm2s)[2:14])
x11();
#pdf("~/Rworks/heart/step2.pdf")
barplot(coef2step[c(3:57,59:68)], xlab = "Variable ID", ylab = "Coefficients Value", main = "Stepwise GLM(BIC)")
axis(side = 1, at = c(1:65)*65/54-0.5, labels = c(3:57,59:68))
#dev.off()