Last data update: 2014.03.03
|
R: Creates the projection pursuit function.
Creates the projection pursuit function.
Description
These functions encapsulate everything, that is, the data, the benchmark and the index parameters, needed to compute the projection index.
Usage
pp(r = 0.8, n, data, oth, k)
Arguments
r |
The radius multiplier. Values between 0.5 and 3 seem to work well.
|
n |
Number of Monte-Carlo Evaluations to approximate the integral. Values as low as 25 can be used.
|
data |
The data for which structure needs to be found.
|
oth |
The benchmark dataset.
|
k |
The target dimension.
|
Details
pp is for projection pursuit.
Value
The actual index function, which takes a single matrix argument, and returns the index value for that projection.
Author(s)
Mohit Dayal
Examples
##Exploring structure in the RANDU data
##Or using the MINSTD generator
randu <- as.matrix(randu)
randtoolbox::setSeed(570)
w <- randtoolbox::congruRand(1200)
dim(w) <- c(3, 400)
w <- t(w)
m <- 'geodesic'
a <- 0.50
ranif1 <- pp(r=1, n=50, data=randu, oth=w, k=2)
set.seed(50)
F1 <- basis_random(3)
o1 <- optim(par=F1, fn=ranif1, gr=basis_nearby(), method='SANN',
control=list(fnscale=-1, maxit=100, trace=1))
plot(randu %*% o1$par)
##How accurate are the values?
ranif1hi <- pp(r=1, n=500, data=randu, oth=w, k=2)
ranif1hi(o1$par)
Results
|