diffusionGraph: Construction of diffusion graph

In kazumits/ddhodge: Modeling Latent Flow Structure using Hodge Decomposition

Description

Construct graph from data matrix based on diffusion maps (Coifman, 2015) with adaptive scaling according to local density.

Usage

diffusionGraph(X, roots, k = 11, npc = min(100, dim(X) - 1),
  ndc = 40, s = 1, j = 7, lambda = 1e-04, sigma = NULL)

Arguments

X	data matrix (columns: n samples, rows: p variables)
roots	indices of columns of X or logical vector; start points of diffusion (i.e., the initial density is 1/n at specified n points).
k	integer; number of nearest neighbours to construct initial k-NN graph
npc	integer; number of principal components in calculating euclid distance.
ndc	integer; number of diffusion components in calculating diffusion distance.
s	numeric; time parameter of diffusion maps.
j	integer; use j-th nearest neighbouor in local scaling of sigma
lambda	numeric; ridge penalty in pulling back divergence
sigma	numeric; fixed sigma for isotropic diffusion

References

"Diffusion maps"

"Self-tuning spectral clustering"

kazumits/ddhodge documentation built on Oct. 17, 2019, 2:38 p.m.