Displaying 1 result from an estimated 1 matches for "cutmid".
Did you mean:
cuid
2012 Oct 20
1
Logistic regression/Cut point? predict ??
...ict(model) #this gives me the a+bx data for new error? I
know I can add type= response to get the probab. but only e^x not *e^x/(1+e^x)
*
pi1hat<-length(z[which(z==1)]/length(z)) My cut point is calculated is
the proportion of 0s to 1.
pi0hat<-length(z[which(z==0)]/length(z))
cutmid <- log(pi0hat/pi1hat)
pred<-ifelse(pred>cutmid,1,0) * I am not sure if I need to compare
these two. I think this is an error.
*
accuracy<-length(which(pred==z))/length(z)
accuracy
rocpreds<-prediction(pred,z)
auc<-performance(rocpreds,"auc")@y.values...