Class Clustering

source code

   object --+    
            |    
BaseEstimator --+
                |
               Clustering

Estimator for the clustering problem, i.e. for

    f(C, mu) = sum_{i=1}^K sum_{x in C_i} |x - mu_i|²,

where C = {C_1, ..., C_K} is a set of sets of points, mu_i is the mean of
C_i and |.|² is the squared Euclidean norm.

This loss function is known as the within-cluster sum of squares.

Parameters
----------
K : Positive integer. The number of clusters to find.

algorithm : Currently only the K-means algorithm (Lloyd's algorithm). The
        algorithm that should be used. Should be one of:
            1. KMeans(...)

        Default is KMeans(...).

algorithm_params : A dictionary. The dictionary algorithm_params contains
        parameters that should be set in the algorithm. Passing
        algorithm=MyAlgorithm(**params) is equivalent to passing
        algorithm=MyAlgorithm() and algorithm_params=params. Default
        is an empty dictionary.

Examples
--------
>>> import parsimony.estimators as estimators
>>> import parsimony.algorithms.cluster as cluster
>>> import numpy as np
>>> np.random.seed(1337)
>>>
>>> K = 3
>>> n, p = 150, 2
>>> X = np.vstack((2 * np.random.rand(n / 3, 2) - 2,
...                0.5 * np.random.rand(n / 3, 2),
...                np.hstack([0.5 * np.random.rand(n / 3, 1) - 1,
...                           0.5 * np.random.rand(n / 3, 1)])))
>>> lloyds = cluster.KMeans(K, max_iter=100, repeat=10)
>>> KMeans = estimators.Clustering(K, algorithm=lloyds)
>>> error = KMeans.fit(X).score(X)
>>> print error
27.6675491884
>>>
>>> #import matplotlib.pyplot as plot
>>> #mus = KMeans._means
>>> #plot.plot(X[:, 0], X[:, 1], '*')
>>> #plot.plot(mus[:, 0], mus[:, 1], 'rs')
>>> #plot.show()

Perform prediction using the fitted parameters.

Finds the closest cluster centre to each point. I.e. assigns a class to
each point.

Returns
-------
closest : A list. A list with p elements: The cluster indices.