ave and unrepresented interaction results (drop=FALSE) - causes errors

问题

Take this simple vector x and two grouping variables, g1 and g2

x <- 1:10
g1 <- rep(1:2, each=5)
g2 <- rep(1:3, c(2,3,5))

There are a number of groups in an interaction of these variables that are not represented in this current data.

table(interaction(g1,g2))
#1.1 2.1 1.2 2.2 1.3 2.3 
#  2   0   3   0   0   5

This causes rare errors that can essentially be boiled down to something simple like:

ave(x, list(g1,g2), FUN=function(x) if(length(x)==0) stop() else length(x) )
#Error in FUN(X[[i]], ...) :

Now, I see no reason why this should be the case, given that the result of ave should always just be the same length as x, with unrepresented interaction of groups discarded. This is an easy fix to ave, such as:

ave2 <- function (x, ..., FUN = mean) 
{
    if (missing(...)) 
        x[] <- FUN(x)
    else {
        # the only real edit is the following line
        g <- interaction(..., drop=TRUE)
        split(x, g) <- lapply(split(x, g), FUN)
    }
    x
}

...which then gives the appropriate result.

ave2(x, list(g1,g2), FUN=function(x) if(length(x)==0) stop() else length(x) )
#[1] 2 2 3 3 3 5 5 5 5 5

I guess before I go and suggest the following to a bug tracker, am I overlooking something simple as to why ave should not have drop=TRUE added?

So the question is, is there a circumstance where ave ever needs to know about unrepresented interactions?

回答1:

Not a problem with ave, but rather stop is not designed to return an item of length 0. Try this instead:

 ave(x, list(g1,g2), FUN=function(x) if(length(x)==0) {vector("numeric",0) 
                                                       }else length(x) )

 [1] 2 2 3 3 3 5 5 5 5 5

回答2:

You can pass the drop argument to ave too:

ave(x,list(g1,g2),drop=TRUE, FUN=function(x) if(length(x)==0) stop() else length(x) )
#[1] 2 2 3 3 3 5 5 5 5 5

来源：https://stackoverflow.com/questions/36348329/ave-and-unrepresented-interaction-results-drop-false-causes-errors

标签

interaction