rtemis (version 0.79)

d.LLE: Locally Linear Embedding

Description

Perform LLE decomposition using lle::lle

Usage

d.LLE(x, k = 2, nn = 0, nn.min = 1, nn.max = 20,
  plot.calcnn = FALSE, id = FALSE, iLLE = FALSE, nnk = TRUE,
  reg = 2, v = 0.9, verbose = TRUE, n.cores = 1, ...)

Arguments

x

Input data

k

Integer vector of length 1 or greater. Rank of decomposition

nn

Integer: Number of neighbors. If Set to 0 (default), will use lle::calc_k to estimate optimal number

nn.min

Integer: Minimum n of neighbors to consider in search, used if nn = 0

nn.max

Integer: Maximum n of enighbors to consider in search, used if nn = 0

plot.calcnn

Logical: If TRUE, print plot after estimation of number of neighbors. Default = FALSE

id

Logical: If TRUE, calculate k (the intrinsic dimension)

iLLE

Logical: If TRUE, use the improved LLE algorithm; see Details in lle::lle Notice: It causes warnings for matrix dimensions (check lle code)

nnk

Logical: If TRUE, use k nearest neighbors method; otherwise, epsilon environment neighbourhood will be used

reg

Integer 1, 2, 3: Regularization methods: See lle::lle("reg")

v

Float: Threshold value for intrinsic dimension estimation. Suggestion for noiseless data: .99, for noisy data: .9. Default = .9

verbose

Logical: If TRUE, print messages to screen. Default = TRUE

n.cores

Integer: Number of cores to use. Default = 1. At some point using more than one cores stopped working. The lle package has not been updated since February 2015 - we will switch to a different implementation soon

...

Additional parameters to be passed to LLE::LLE

Value

rtDecom object

Details

Project scaled variables to LLE components Input must be n by p, where n represents number of cases, and p represents number of features. LLE will be applied to the transpose of the n x p matrix.

See Also

Other Decomposition: d.CUR, d.H2OAE, d.H2OGLRM, d.ICA, d.ISOMAP, d.KPCA, d.MDS, d.NMF, d.PCA, d.SPCA, d.SVD, d.TSNE, d.UMAP