calc_k: Calculation of the optimal number of neighbours.

In lle: Locally linear embedding

Description

Calculates of optimal number of neighbours by using the algorithm proposed by Kayo (see Ref.). Therefore the LLE alorithm is performed for every k-neighbourhood size.

Usage

 calc_k(X, m, kmin=1, kmax=20, plotres=TRUE, 
                  parallel=FALSE, cpus=2, iLLE=FALSE) 

Arguments

X	matrix object containing the input data.
m	intrinsic dimension of the data.
kmin	minimal value of k.
kmax	maximal value of k.
plotres	a logical values indicating wheather to plot the result.
parallel	a logical values indicating wheather to use parallel computation on multiple cpu cores. See snowfall.
cpus	number of cpus cores used for parallel computation.
iLLE	a logical values indicating wheater to use improved LLE (very CPU intensive). See lle.

Details

Since the calculation of the optimal number of neighbours m is a step that is normally applied before the execution of LLE itself, the intrinsic dimension may be unknown. In this case, a good guess is sufficient. If no good estimation can be made the largest plausible value should be chosen.

Value

res	dataframe containing the number of neighbours and the calculated parameter ρ. The number of neighbours belonging to the smallest value of ρ should be chosen.

References

Locally linear embedding algorithm - extensions and applications / Olga Kayo / Universitatis Ouluensis, Oulu, Finland / 2006

Examples

 
   ## Not run: 
	data( lle_scurve_data )
	X <- lle_scurve_data
	calc_k( X, 2, 1, 15 ) 

	data( lle_scurve_data )
	X <- lle_scurve_data
	calc_k( X, 2, 1, 15, FALSE, TRUE, 4 ) 
   
## End(Not run)

lle documentation built on May 2, 2019, 2:49 p.m.