FOSC

A Python version of the Framework for optimal extraction of clusters (FOSC) (CAMPELLO et al., 2013) [1].

Note: This is a test version and may contain bugs. If you notice anything wrong, please contact the developers.

Quick Start Tutorial

This is a tutorial that teaches how to quickly start using the FOSC package.

Considering you have a dataset in a CSV file, start by loading its data.

import numpy as np

mat = np.genfromtxt(file_path, dtype=float, delimiter=',', usecols=range(0, -1), missing_values=np.nan)

Note:

Replace file_path with your dataset's path.

In this example, the last column represents the cluster to which an object is assigned, according to another algorithm. Since we are using our own method, we ignore the last column with the usecols parameter.

Choose the MClSize (minimum cluster size) parameter and the method of linkage:

mclsize = 30
linkage_method = "single"

Calculate the distance matrix:

from scipy.spatial import distance_matrix

dist_mat = distance_matrix(mat, mat, p=2)

Now, you can instantiate the FOSC object:

from FOSC import FOSC

foscFramework = FOSC(dist_mat, linkage_method, mclsize)

And run the FOSC algorithm:

infiniteStability = foscFramework.propagateTree()
partition, lastObjects = foscFramework.findProminentClusters(1, infiniteStability)

The partition returned from the findProminentClusters method is the result of running the FOSC algorithm, i.e., the cluster extraction.

Finally, you can visualize the results using the functions available in the Plot module:

from FOSC import Plot

Plot.plotDendrogram(foscFramework)
Plot.plotSilhouette(foscFramework, partition)
Plot.plotReachability(foscFramework, partition)

References

[1] CAMPELLO, R. J.; MOULAVI, D.; ZIMEK, A.; SANDER, J. A framework for semi-supervised and unsupervised optimal extraction of clusters from hierarchies. Data Mining and Knowledge Discovery, Springer, v. 27, p. 344–371, 2013.