PCR vs. PLS (part 7 - cont(a))

This post continue with the plots we have seen in PCR vs. PLS (part 7) where we saw the cross validation error for the training samples (odd samples from the soy meal), with the models developed with PCR and PLS.

 

Now we want to check how these models performs with the even samples that we have place apart in a test set.

 

Now we overlap the RMSEP statistics for the test set  (for PLS and PCR models).

#from: PCR vs. PLS (part 7)
## add extra space to right margin of plot within frame
par(mar=c(5, 4, 4, 6) + 0.1)
plot(Xodd_pcr3,"validation",estimate="CV",ylim=c(0.5,2.0),

     xlim=c(1,10),col="blue",lwd=2,
     main=" Corr & RMSEP PCR vs PLS")
par(new=TRUE)
plot(Xodd_pls3,"validation", estimate="CV",

ylim=c(0.5,2.0),xlim=c(1,10),col="red",lwd=2,
     main="")

 ## New code added

rmsep_pls3_even_10t<-RMSEP(Xodd_pls3,ncomp=c(1:10),
                           newdata=soy_ift_even,
                           intercept=FALSE)
rmsep_pcr3_even_10t<-RMSEP(Xodd_pcr3,ncomp=c(1:10),
                           newdata=soy_ift_even,
                           intercept=FALSE)
par(new=TRUE)
plot(rmsep_pls3_even_10t,col="red",
     lwd=2,ylim=c(0.5,2.0),
     lty=5,axes=FALSE,type="l",
     xlab="",ylab="",main="")
par(new=TRUE)
plot(rmsep_pcr3_even_10t,col="blue",
     lwd=2,ylim=c(0.5,2.0),
     lty=5,axes=FALSE,type="l",
     xlab="",ylab="",main="")
legend("bottomleft",legend=c("PCR-RMSECV train set",
       "PLS-RMSECV train set","PCR-RMSEP even set",
       "PLS-RMSEP even set"),
       col=c("blue","red","blue","red"),
       lty=c(1,1,5,5),lwd=1)

As we can see the Model with PLS performs better for the Even sample set and also for the cross validation.