svcm (version 0.1.2)

resolution: Re-scaling resolution of SVCM predictors and effects

Description

This routine serves post-hoc adjustment of the resolution of a space-varying coefficient model estimated by svcm.

Usage

resolution(X, svcmlist, fac)

Arguments

X

(r x p)-array of covariates

svcmlist

return list of function svcm

fac

2d or 3d vector of scaling factors

Value

A list with components

fitted

fitted values at fac-times rescaled resolution.

effects

estimated effects at fac-times rescaled resolution.

Details

The basis function approach underlying svcm allows to rescale the original resolution by evaluating the basis functions at additional points. Assuming that the voxel center is most representative for the whole voxel, fac-times resolution of 1d data with $n$ voxels sized $vsize$ bases on coordinates $$\left(i - \frac{1}{2}\right) \cdot \frac{vsize}{fac} , \quad i = 1, \ldots, n \cdot fac.$$ See also doc file resolution\_scheme.pdf. The formula is applied into x-, y- and z-direction and results in a refined equidistant 2d resp. 3d grid. It also means that, in general, the resized arrays of predictors and effects do no longer contain the values at the former coordinates.

Note that memory requirements can be enormous depending on object size and the intended resolution.

Examples

##DTI data; regressors are given by the diffusion weigthing gradients
data(brain2d)
X <- matrix(c(0.5, 0.5,   0,   0, 0.5, 0.5,
                0,   0, 0.5, 0.5, 0.5, 0.5,
              0.5, 0.5, 0.5, 0.5,   0,   0,
                0,   0,   0,   0,   1,  -1,
                1,  -1,   0,   0,   0,   0,
                0,   0,   1,  -1,   0,   0), nrow = 6)
M <- svcm(brain2d, X, knots=c(60, 50), deg=c(1, 1), vsize=c(1.875,
          1.875), search=TRUE, type="SEQ", lambda.init=rep(0.1, 2),
          lower=rep(-5, 2), upper=rep(0, 2), ngrid=10)
M2 <- resolution(X, M, fac=c(2, 2))

##show data extract at original and double resolution
extract <- list(M$fit[21:40, 21:60, 1],
                M2$fit[(21*2):(40*2), (21*2):(60*2), 1], 
                M$eff[21:40, 21:60, 1],
                M2$eff[(21*2):(40*2), (21*2):(60*2), 1]) 
zlim1 <- range(extract[[1]], extract[[2]])
zlim2 <- range(extract[[3]], extract[[4]])
old.par <- par(no.readonly = TRUE) 
par(pin=c(3*1, 3*0.67), mfrow=c(2, 2)) 
image(t(extract[[1]]), axes=FALSE, zlim=zlim1, col=grey.colors(256))
title("Fitted Values")
image(t(extract[[2]]), axes=FALSE, zlim=zlim1, col=grey.colors(256))
title("Fitted Values at Double Resolution")
image(t(extract[[3]]), axes=FALSE, zlim=zlim2, col=grey.colors(256))
title("Estimated VC Surface (1st DT element)")
image(t(extract[[4]]), axes=FALSE, zlim=zlim2, col=grey.colors(256))
title("VC Surface at Double Resolution")
par(old.par)