GFQ: Greedy Functional Quantization

In elamrireda/FunctQuant: Data-Driven Functional Quantization

Description

Data-driven greedy functional quantization based on the distorsion error or the maximin (a space filling design criterion)

Usage

GFQ(data,mKL,size,method,deepstart=TRUE)

Arguments

data	matrix that we want to quantize.
mKL	truncation argument for the dimension reduction.
size	size of the quantization grids.
method	"L2" or "maximin".
deepstart	maximin, if TRUE: the quantization is started by the central curve.

Value

data	the input matrix.
quantizer	the quantizer grid (curves are chosen among the input data.
weights	the associated weight of each curve (calculated using the input matrix).

See Also

CVT and StochGradient

Examples

##### function to generate realizations of BM
BM <- function(N=1000,M=1,x0=0,t0=0,T=1,Dt=NULL)
{
  Dt <- (T - t0)/N
  t <- seq(t0, T, by=Dt)
  res <- data.frame(sapply(1:M,function(i) c(0,cumsum(rnorm(N,mean  =0,sd=sqrt(Dt))))))
  names(res) <- paste("X",1:M,sep="")
  X <- ts(res, start = t0, deltat = Dt)
  return(X)
}

data <- t(BM(N=200-1,M=200))
mKL <- 2
size <- 10
method <- "maximin"
quant <- GFQ(data,mKL,size,method,deepstart=TRUE)

elamrireda/FunctQuant documentation built on Jan. 1, 2021, 2:50 p.m.