src/bic.h File Reference

Template function implementations of the Bayesian Information Criterion. More...

#include <stdint.h>
#include <numeric>
#include <iostream>
#include <cmath>
#include <vector>
#include "dissimilarity.h"
#include "partition.h"

Include dependency graph for bic.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Namespaces
namespace	cluster
	Namespace for everything in the cluster library.
Functions
template<typename D >
double	bic (const partition &p, D distance, size_t M)
	Directly computes the BIC from a partition object based on the cluster centroids and the number of clusters.
template<typename SizeIterator , typename DissimIterator >
double	bic (size_t k, SizeIterator cluster_sizes, DissimIterator sum2_dissim, size_t dimensionality)
	This version of the BIC assumes some precomputed information.

---

Detailed Description

Template function implementations of the Bayesian Information Criterion.

Author:

Todd Gamblin tgamblin@llnl.gov The Bayesian Information Criterion (BIC) is a criterion for model selection that balances a maximum likelihood estimator and a parameter count. This implementation is designed for clustering algorithms, in particular K-means and K-medoids clustering algorithms where we expect clusters with spherical gaussian distributions.

Here, we want to test whether a clustering's centroids or medoids are good predictors of the points in a data set, so these are our parameters, and we try to find the best clustering without too many clusters. For more on this technique and the approach we've based this implementation on, see this paper:

Dan Pelleg and Andrew Moore. X-Means: Extending K-Means with Efficient Estimation of the Number of Clusters. Proceedings of the Seventeenth International Conference on Machine Learning, San Francisco, CA. June 29-July 2, 2000. pp 727-734.

Definition in file bic.h.