modelimage: Model selection image

In rdetools: Relevant Dimension Estimation (RDE) in Feature Spaces

Description

The function produces a graphical illustration of a model selection which has been done with selectmodel. Strictly speaking it's a filled.contour plot in which additionally the relevant dimensions for the different models are drawn as a black line. selectmodel chooses the deepest point in this map, that is the model and the relevant dimension with the smallest loo-cv-error/negative-log-likelihood-value.

Usage

modelimage(model,
           color.palette = topo.colors,
	   log = TRUE,
	   plottitle = "RDE Model Selection",
	   ...)

Arguments

model	list of model selection data as it has been returned by selectmodel
color.palette	color palette function to use, see rainbow
log	leave this TRUE, if the axis of the model parameter should be logarithmically scaled. Set this to FALSE if you want linear scaling.
plottitle	title of the plot
...	additional parameters for filled.contour

Author(s)

Jan Saputra Mueller

References

M. L. Braun, J. M. Buhmann, K. R. Mueller (2008) \_On Relevant Dimensions in Kernel Feature Spaces\_

See Also

selectmodel, distimage, drawkpc, filled.contour, rainbow

Examples

## model selection with RBF-kernel and graphical illustration
d <- sincdata(100, 0.1) # generate sinc data
# do model selection
m <- selectmodel(d$X, d$y, sigma = logspace(-3, 3, 100))
modelimage(m) # draw model selection image

rdetools documentation built on May 2, 2019, 7:02 a.m.