simest (version 0.4-1-1)

fastmerge: Pre-binning of Data Points

Description

Numerical tolerance problems in non-parametric regression makes it necessary for pre-binning of data points. This procedure is implicitly performed by most of the regression functions in R. This function implements this procedure with a given tolerance level.

Usage

fastmerge(DataMat, w = NULL, tol = 1e-4)

Value

A list including the elements

DataMat

a numeric matrix/vector with rows sorted and possibly merged with respect to the first column.

w

obtained weights corresponding to the merged points.

Arguments

DataMat

a numeric matrix/vector with rows as data points.

w

an optional numeric vector of the same length as \(x\); defaults to all elements 1.

tol

a numeric value providing the tolerance for identifying duplicates with respect to the first column DataMat[,1].

Author

Arun Kumar Kuchibhotla; also the authors of smooth.spline.

Details

If two values in the first column of DataMat are separated by a value less than tol then the corresponding rows are merged.

See Also

The function smooth.spline also uses such pre-binning.

Examples

 ## relevant example % found in ../tests/fastmerge-ex.R
n <- 47
set.seed(2657) # <- found after searching
x <- sort(signif(runif(n, -1,1), 5))
y <- sinpi(3*x) * exp(-x) + rnorm(n)/10
str(fmL <- fastmerge(cbind(x,y))) # only 44 (out of 47) "unique" x[]
d.fm <- data.frame(fmL)
d2 <- data.frame(fastmerge(cbind(x,y), tol = 25e-4)) # larger tol ==> only 42 "unique"
table(w <- d2$w) # 3x w=2  and  1 w=3
stopifnot(nrow(d.fm) == 44, nrow(d2) == 42,
          identical(    w[w > 1], c(2, 2, 2, 3)),
          identical(which(w > 1), c(5L, 26L, 28L, 39L)),
          all.equal(1000 * fmL$AddVar[fmL$w != 1],
                    c(2.28919, 23.918, 17.5813), tolerance = 3e-6))

plot(y ~ x, type = "b")
lines(d.fm[,1], d.fm[,2], col = adjustcolor(2, 1/2), lwd=3)
lines(d2  [,1], d2  [,2], col = adjustcolor(4, 1/2), lwd=2)
abline(v = d.fm[d.fm$w > 1, 1],  col = 2, lwd=3, lty=2)
abline(v = (xw <- d2[w > 1, 1]), col = 4, lwd=2, lty=3)
axis(3, at= xw, labels=paste("w=",w[w > 1]), col = 4, col.axis = 4)