R help - Nov 2009 - 3-D Plotting of predictions from GAM/GAMM object

Hello all,
  Thank you for the previous assistance I received from this listserve!
  My current question is: How can I create an appropriate matrix of 
values from a GAM (actually a GAMM) to make a 3-D plot? This model is 
fit as a tensor product spline of two predictors and I have used it to 
make specific predictions by calling:

YoyRasPred6<-predict(YoyRas.Distribution.T.L.DT.gamm$gam,newdata=YoyRasSubset6,se.fit=T)

However, this type of command, I believe, produces only a vector of 
values. I have plotted these values and know how to do so in 2-D. 
However, I would like to create a 3-D plot, which I believe will require 
an entire matrix of predictions (z values) based on all possible inputs 
(x and y values). What is an efficient way to do this? It seems there 
must be a better way than simply predicting vector after vector ... as I 
am now, then piecing these together.

Thank you very much for any answers or assistance in general! Best wishes.

Sincerely,
Paul Simonin

On 21/11/2009 3:11 PM, Paul Simonin wrote:
> Hello all,
>   Thank you for the previous assistance I received from this listserve!
>   My current question is: How can I create an appropriate matrix of 
> values from a GAM (actually a GAMM) to make a 3-D plot? This model is 
> fit as a tensor product spline of two predictors and I have used it to 
> make specific predictions by calling:
> 
>
YoyRasPred6<-predict(YoyRas.Distribution.T.L.DT.gamm$gam,newdata=YoyRasSubset6,se.fit=T)
> 
> However, this type of command, I believe, produces only a vector of 
> values. I have plotted these values and know how to do so in 2-D. 
> However, I would like to create a 3-D plot, which I believe will require 
> an entire matrix of predictions (z values) based on all possible inputs 
> (x and y values). What is an efficient way to do this? It seems there 
> must be a better way than simply predicting vector after vector ... as I 
> am now, then piecing these together.
You can use expand.grid() to generate a dataframe corresponding to the 
grid of values, then reshape the results into a matrix.   I don't have 
your data to show a fit, but this shows how the graphing is done.

 > x <- 1:10
 > y <- 1:12
 > grid <- expand.grid(x=x, y=y)  # give names to get named columns
 > z <- matrix(apply(grid, 1, prod), nrow=10) # use predict instead here
 > library(rgl)
 > persp3d(x,y,z, col="red")

Duncan Murdoch