Question

In cor function , there is a wonderful argument the ” use “(e.g. : use=”na.or.complete” )

I tried to use the same argument with calculating RMSE or bias. It seems that it is not working.I wonder what we could use instead.

RMSE <- function(x, y){
sqrt(mean((x-y)^2))}
RMSE(x,y, use = "na.or.complete")
Error in RMSE(x,y, use = "na.or.complete") : 
unused argument(s) (use = "na.or.complete")

Answer 1

The mean function to compute THE RMSE has an na.rm argument, You can use it like this :

RMSE <- function (pred,obs,na.rm=FALSE){
   sqrt(mean((pred - obs)^2, na.rm = na.rm))
}

The na.rm argument can be used to mimic na.or.complete behavior since in almost cases , using na.or.complete option is equiavlent to remove NA value from the original pairs. For example :

swM <- swiss[,1:2]
swM[1,2] <-  NA 
swM[1,1] <-  NA 
identical(cor(swM, use = "na.or.complete"),
          cor(na.exclude(swM), use = "all"))
TRUE

With RMSE

swM[1,2] <-  NA 
swM[1,1] <-  NA 
identical(RMSE(swM[,1],swM[,2],na.rm=T),
          RMSE(swM[,1],swM[,2],na.rm=F))
 FALSE

EDIT set very small values to NA

I don’know how you store your data. But I assume you have a data.table with a variable x. You can do something like this :

 dat$x[dat$x <= -3.4e+38] <- NA