[R] Adding regression surface to cloud plot

[Deepayan Sarkar]

On Monday 26 April 2004 08:14, Anne wrote:
> Hello!
> I would like to add a plot of the regression surface to my cloud plot
> . Is it possible? Thanks

Possible (since 1.9.0), but not easy. Technically, the biggest problem 
is deciding which points are 'behind' the surface and which 'in 
front' (that would decide the order of plotting). Otherwise, all the 
pieces are in place.

The following should work for regression surfaces, but may need an 
inequality to be changed for specific viewpoints:


panel.3dreg <-
    function(x, y, z, rot.mat, distance,
             xlim.scaled, ylim.scaled, 
             nmesh = 21, ...)
{
    fm <- lm(z ~ x + y)
    id <- predict(fm) < z ## may need to be reversed
    panel.3dscatter(x[id], y[id], z[id], rot.mat, distance,
                    xlim.scaled = xlim.scaled,
                    ylim.scaled = ylim.scaled,
                    ...)
    x.locs <- seq(xlim.scaled[1], xlim.scaled[2], length = nmesh)
    y.locs <- seq(ylim.scaled[1], ylim.scaled[2], length = nmesh)
    grid <-
        expand.grid(y = y.locs, x = x.locs)
    grid$z <- predict(fm, newdata = grid)
    panel.3dwire(x = x.locs, y = y.locs, z = grid$z,
                 rot.mat, distance,
                 xlim.scaled = xlim.scaled,
                 ylim.scaled = ylim.scaled,
                 ..., col.at = .5, col.regions = "transparent")
    panel.3dscatter(x[!id], y[!id], z[!id], rot.mat, distance,
                    xlim.scaled = xlim.scaled,
                    ylim.scaled = ylim.scaled,
                    ...)
}

## example
data(rock)
cloud(area ~ peri * shape, rock,
      panel.3d.cloud = panel.3dreg,
      nmesh = 40, pch = 16)


- Deepayan