-
Notifications
You must be signed in to change notification settings - Fork 0
/
code function to compare model error, bias etc.R
47 lines (47 loc) · 2.61 KB
/
code function to compare model error, bias etc.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
#-------------------------------------------------------------------------------------------
#function to compare model error, bias etc.
#-------------------------------------------------------------------------------------------
library(ncvreg)
penalty=function(s,gamma_MCP,gamma_SCAD){
Lasso_data=matrix(0,nrow=50,ncol=7)
MCP_data=matrix(0,nrow=50,ncol=7)
SCAD_data=matrix(0,nrow=50,ncol=7)
for(i in 1:50){
X=matrix(rnorm(n*p), nrow=n, ncol=p)%*%SigmaXsqrt
y=X%*%beta + rnorm(n, sd = s)
cv_fit_Lasso=cv.glmnet(X, y, alpha = 1, lambda = lambda_values)
lasso_opt_lambda=cv_fit_Lasso$lambda.min
lasso_reg=glmnet(X, y,alpha = 1,lambda=lasso_opt_lambda)
beta_hat=matrix(coef(lasso_reg),nrow = 301)
beta_hat=beta_hat[-1,]
Lasso_data[i,1]=t(beta_hat-beta)%*%SigmaX%*%(beta_hat-beta)
Lasso_data[i,2:5]=c(abs(beta_hat[1]-beta[1]), abs(beta_hat[3]-beta[3]),
abs(beta_hat[5]-beta[5]), abs(beta_hat[6]-beta[6]))
Lasso_data[i,6]=length(which(beta_hat!=0 & beta!=0))/length(which(beta!=0))
Lasso_data[i,7]=length(which(beta_hat==0 & beta==0))/length(which(beta==0))
cv_fit_MCP=cv.ncvreg(X, y, alpha = 1, lambda = lambda_values)
MCP_opt_lambda=cv_fit_MCP$lambda.min
MCP_reg=ncvreg(X, y,alpha = 1,lambda=MCP_opt_lambda,type="MCP", gamma = gamma_MCP)
beta_hat=matrix(coef(MCP_reg),nrow=301)
beta_hat=beta_hat[-1,]
MCP_data[i,1]=t(beta_hat-beta)%*%SigmaX%*%(beta_hat-beta)
MCP_data[i,2:5]=c(abs(beta_hat[1]-beta[1]),abs(beta_hat[3]-beta[3]),
abs(beta_hat[5]-beta[5]), abs(beta_hat[6]-beta[6]))
MCP_data[i,6]=length(which(beta_hat!=0 & beta!=0))/length(which(beta!=0))
MCP_data[i,7]=length(which(beta_hat==0 & beta==0))/length(which(beta==0))
cv_fit_SCAD=cv.ncvreg(X, y, alpha = 1, lambda = lambda_values)
SCAD_opt_lambda=cv_fit_SCAD$lambda.min
SCAD_reg=ncvreg(X, y,alpha = 1,lambda=SCAD_opt_lambda,type="SCAD", gamma = gamma_SCAD)
beta_hat=matrix(coef(SCAD_reg),nrow=301)
beta_hat=beta_hat[-1,]
SCAD_data[i,1]=t(beta_hat-beta)%*%SigmaX%*%(beta_hat-beta)
SCAD_data[i,2:5]=c(abs(beta_hat[1]-beta[1]),abs(beta_hat[3]-beta[3]),
abs(beta_hat[5]-beta[5]), abs(beta_hat[6]-beta[6]))
SCAD_data[i,6]=length(which(beta_hat!=0 & beta!=0))/length(which(beta!=0))
SCAD_data[i,7]=length(which(beta_hat==0 & beta==0))/length(which(beta==0))
}
mean_data=cbind(colMeans(Lasso_data),colMeans(MCP_data),colMeans(SCAD_data))
rownames(mean_data)=c("Error","Bias1","Bias3","Bias5","Bias6","TPR","TNR")
colnames(mean_data)=c("LASSO","MCP","SCAD")
return(mean_data)
}