[Rd] interp.surface() error (PR#7745)

[murdoch at stats.uwo.ca]

On Wed, 23 Mar 2005 22:16:30 +0100 (CET), takahama at andrew.cmu.edu
wrote :

>R version: 2.001
>OS: Windows XP SP2
>
>When I use the interp.surface() function, on occasion I have encountered 
>the following error message:
>
>Error in interp.surface(a, loc) : subscript out of bounds
>
>Since it is somewhat dependent on the data set, I cannot say exactly in 
>which cases these occur, but I believe I have found the problem and a 
>solution for it:
>
>The scaling calculations in the code,
>    lx <- ((nx - 1) * (loc[, 1] - xa))/xr + 1
>    ly <- ((ny - 1) * (loc[, 2] - ya))/yr + 1
>can lead to numerical error, so in the following lines
>    lx1 <- ifelse(lx == nx, nx - 1, trunc(lx))
>    ly1 <- ifelse(ly == ny, ny - 1, trunc(ly))
>the equality statements which are the first arguments to the IF-ELSE 
>statements return false when they are supposed to be true. I believe the 
>remedy is to simply avoid absolute equalities, as in the following example 
>(which has worked for my problem):
>    lx1 <- ifelse(abs(lx - nx) < 1e-10, nx - 1, trunc(lx))
>    ly1 <- ifelse(abs(ly - ny) < 1e-10, ny - 1, trunc(ly))
>
>I hope this is an appropriate posting for an error report. Thank you very 
>much for your hard work,

I think 

ifelse(lx >= nx, nx - 1, trunc(lx))
ifelse(ly >= ny, ny - 1, trunc(ly))

would probably be a better fix, but I'd really like to see an example
of this in action.  Could you email me a dataset where you see it?

Duncan Murdoch