"JarvisPatrick" (Machine Learning Method)

Details & Suboptions

  • "JarvisPatrick" is a neighbor-based clustering method. "JarvisPatrick" works for arbitrary cluster shapes and sizes. However, it is parameter sensitive, and can fail when clusters have different densities or are loosely connected.
  • The following plots show the results of the "JarvisPatrick" method applied to toy datasets:
  • The algorithm finds clusters based on the similarity of nearest neighbors of data points, and uses "shared nearest neighbors" as a measure of similarity between points.
  • In "JarvisPatrick", neighbors are defined by points within a ball of ϵ radius. Each couple of neighbors that share at least p neighbors belong to the same cluster.
  • The following suboptions can be given:
  • "NeighborhoodRadius" Automaticradius ϵ
    "SharedNeighborsNumber" Automaticminimum number of shared neighbors p

Examples

open allclose all

Basic Examples  (2)

Find clusters of nearby values using the "JarvisPatrick" method:

Create random 2D vectors:

Plot clusters in data found using the "JarvisPatrick" method:

Options  (5)

DistanceFunction  (1)

Find the clustering of the strings using the edit distance and specifying the "NeighborhoodRadius":

"NeighborhoodRadius"  (2)

Find clusters using "JarvisPatrick":

Find clusters by specifying the "NeighborhoodRadius" suboption:

Define a set of two-dimensional data points, characterized by four somewhat nebulous clusters:

Plot different clusterings of the data using the "JarvisPatrick" method by varying the "NeighborhoodRadius":

"SharedNeighborsNumber"  (2)

Find clusters by varying "NeighborsNumber" suboptions:

Create random 2D vectors:

Plot clustering of data points using the "JarvisPatrick" method by varying the "SharedNeighborsNumber":

Applications  (1)

Create and visualize noisy 2D moon-shaped training and test datasets:

Train various ClassifierFunctions by varying the "NeighborhoodRadius" using the "JarvisPatrick" method:

Find and visualize different clusterings of the test set, given small changes in "NeighborhoodRadius":