lv_optics: lv_optics

In elbamos/largeVis: High-Quality Visualizations of Large, High-Dimensional Datasets

Description

Experimental implementation of the OPTICS algorithm.

Usage

lv_optics(edges, neighbors, eps = Inf, minPts = nrow(neighbors), eps_cl, xi,
  useQueue = TRUE, verbose = getOption("verbose", TRUE))

Arguments

edges	A weighted graph of the type produced by buildEdgeMatrix. Alternatively, a largeVis object, in which case edges and neighbors will be taken from the edges and knns parameters, respectively.
neighbors	An adjacency matrix of the type produced by randomProjectionTreeSearch
eps	See optics.
minPts	See optics.
eps_cl	See optics.
xi	See optics.
useQueue	Whether to process points in order of core distance. (See note.)
verbose	Vebosity level.

Details

This is an implementation of the OPTICS algorithm that attempts to leverage the largeVis nearest-neighbor search.

This implementation does not use the OPTICS neighbor-search strategy, in favor of using the pre-calculated neighbor matrix produced incidentally by largeVis. It is therefore a variant of OPTICS rather than an implementation of the original, and the results vary slightly from those obtained by the implementations in ELKI and the dbscan package.

Value

An optics object.

Note

The useQueue parameter controls the order in which points that have not yet been visisted are processed. If FALSE, points are processed in order of rows. If TRUE, they are processed in ascending order of core distance. FALSE is more compatible with the implementations in the dbscan package and in the ELKI Java clustering package. TRUE may produce preferrable results.

References

Mihael Ankerst, Markus M. Breunig, Hans-Peter Kriegel, Jorg Sander (1999). OPTICS: Ordering Points To Identify the Clustering Structure. ACM SIGMOD international conference on Management of data. ACM Press. pp. 49-60.

elbamos/largeVis documentation built on May 16, 2019, 2:58 a.m.