smile.clustering

Class SIB

java.lang.Object

smile.clustering.PartitionClustering<double[]>

smile.clustering.SIB

All Implemented Interfaces:

Clustering<double[]>

public class SIB
extends PartitionClustering<double[]>

The Sequential Information Bottleneck algorithm. SIB clusters co-occurrence data such as text documents vs words. SIB is guaranteed to converge to a local maximum of the information. Moreover, the time and space complexity are significantly improved in contrast to the agglomerative IB algorithm.

In analogy to K-Means, SIB's update formulas are essentially same as the EM algorithm for estimating finite Gaussian mixture model by replacing regular Euclidean distance with Kullback-Leibler divergence, which is clearly a better dissimilarity measure for co-occurrence data. However, the common batch updating rule (assigning all instances to nearest centroids and then updating centroids) of K-Means won't work in SIB, which has to work in a sequential way (reassigning (if better) each instance then immediately update related centroids). It might be because K-L divergence is very sensitive and the centroids may be significantly changed in each iteration in batch updating rule.

Note that this implementation has a little difference from the original paper, in which a weighted Jensen-Shannon divergence is employed as a criterion to assign a randomly-picked sample to a different cluster. However, this doesn't work well in some cases as we experienced probably because the weighted JS divergence gives too much weight to clusters which is much larger than a single sample. In this implementation, we instead use the regular/unweighted Jensen-Shannon divergence.

References

1. N. Tishby, F.C. Pereira, and W. Bialek. The information bottleneck method. 1999.
2. N. Slonim, N. Friedman, and N. Tishby. Unsupervised document classification using sequential information maximization. ACM SIGIR, 2002.
3. Jaakko Peltonen, Janne Sinkkonen, and Samuel Kaski. Sequential information bottleneck for finite data. ICML, 2004.

Author:

Haifeng Li

Field Summary

Fields inherited from interface smile.clustering.Clustering

OUTLIER

Constructor Summary

Constructors

Constructor and Description
SIB(double[][] data, int k) Constructor.
SIB(double[][] data, int k, int maxIter) Constructor.
SIB(double[][] data, int k, int maxIter, int runs) Constructor.
SIB(SparseDataset data, int k) Constructor.
SIB(SparseDataset data, int k, int maxIter) Constructor.
SIB(SparseDataset data, int k, int maxIter, int runs) Constructor.

Method Summary

Methods

Modifier and Type	Method and Description
double[][]	centroids() Returns the centroids.
double	distortion() Returns the distortion.
int	predict(double[] x) Cluster a new instance.
int	predict(SparseArray x) Cluster a new instance.
String	toString()

Methods inherited from class smile.clustering.PartitionClustering

getClusterLabel, getClusterSize, getNumClusters

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Detail

SIB

public SIB(double[][] data,
   int k)

Constructor. Clustering data into k clusters up to 100 iterations.

Parameters:

data - the normalized co-occurrence input data of which each row is a sample with sum 1.

k - the number of clusters.

SIB

public SIB(double[][] data,
   int k,
   int maxIter)

Constructor. Clustering data into k clusters.

Parameters:

data - the input data of which each row is a sample.

k - the number of clusters.

maxIter - the maximum number of iterations.

SIB

public SIB(double[][] data,
   int k,
   int maxIter,
   int runs)

Constructor. Run SIB clustering algorithm multiple times and return the best one.

Parameters:

data - the input data of which each row is a sample.

k - the number of clusters.

maxIter - the maximum number of iterations.

runs - the number of runs of SIB algorithm.

SIB

public SIB(SparseDataset data,
   int k)

Constructor. Clustering data into k clusters up to 100 iterations.

Parameters:

data - the sparse normalized co-occurrence dataset of which each row is a sample with sum 1.

k - the number of clusters.

SIB

public SIB(SparseDataset data,
   int k,
   int maxIter)

Constructor. Clustering data into k clusters.

Parameters:

data - the sparse normalized co-occurrence dataset of which each row is a sample with sum 1.

k - the number of clusters.

maxIter - the maximum number of iterations.

SIB

public SIB(SparseDataset data,
   int k,
   int maxIter,
   int runs)

Constructor. Run SIB clustering algorithm multiple times and return the best one.

Parameters:

data - the sparse normalized co-occurrence dataset of which each row is a sample with sum 1.

k - the number of clusters.

maxIter - the maximum number of iterations.

runs - the number of runs of SIB algorithm.

Method Detail

predict

public int predict(double[] x)

Cluster a new instance.

Parameters:

x - a new instance.

Returns:

the cluster label.

predict

public int predict(SparseArray x)

Cluster a new instance.

Parameters:

x - a new instance.

Returns:

the cluster label.

distortion

public double distortion()

Returns the distortion.

centroids

public double[][] centroids()

Returns the centroids.

toString

public String toString()

Overrides:

toString in class Object