scmsMParzen: MParzen KDE-based Subspace Constrained Mean Shift

In rudazhang/plmr: Probabilistic Learning on Manifolds in R

Description

Using the log Manifold Parzen windows (MParzen) Gaussian KDE.

Usage

scmsMParzen(Y0, X, model, sigma, h, r = NULL, minCos = 0.01, maxIter = 100L)

Arguments

Y0	Initial points, an n-by-M matrix.
X	Data, an n-by-N matrix.
model	Top d-th order local covariance structure; output of 'MParzenTrain()'.
sigma	Maximum noise level added to the normal space of each kernel, i.e. marginal standard deviation of an isotropic Gaussian. Can be zero.
h	Bandwidth for isotropic Gaussian kernel density estimation, a scalar. Determines step size.
r	Radius of nearest neighbors to use in evaluating density and derivatives. Default to use all data.
minCos	Convergence criterion, cosine of the angle between gradient and its component in the "local normal space".
maxIter	Maximum number of iterations.

Value

A list: 'Y', the final points; 'cosNormal', convergence score; 'updatedPoints', the number of updated points per update.

rudazhang/plmr documentation built on Aug. 30, 2021, 5:43 p.m.