スペクトラルクラスタリング

# ライブラリの読み込み

print(__doc__)

import time

import warnings

import numpy as np

import matplotlib.pyplot as plt

from sklearn import cluster, datasets, mixture

from sklearn.preprocessing import StandardScaler

from itertools import cycle, islice

np.random.seed(0)

# データセット作成

n_samples = 1500

noisy_circles = datasets.make_circles(n_samples=n_samples, factor=.5,

                                      noise=.05)

noisy_moons = datasets.make_moons(n_samples=n_samples, noise=.05)

# クラスターのパラメーター作成

plt.figure(figsize=(9 * 2 + 3, 12.5))

plt.subplots_adjust(left=.02, right=.98, bottom=.001, top=.96, wspace=.05,

                    hspace=.01)

plot_num = 1

default_base = {‘quantile’: .3,

                ‘eps’: .3,

                ‘damping’: .9,

                ‘preference’: -200,

                ‘n_neighbors’: 10,

                ‘n_clusters’: 3}

datasets = [

    (noisy_circles, {‘damping’: .77, ‘preference’: -240,

                     ‘quantile’: .2, ‘n_clusters’: 2}),

    (noisy_moons, {‘damping’: .75, ‘preference’: -220, ‘n_clusters’: 2}),]

for i_dataset, (dataset, algo_params) in enumerate(datasets):

    params = default_base.copy()

    params.update(algo_params)

    X, y = dataset

    #データセットを正規化する

    X = StandardScaler().fit_transform(X)

    #帯域幅作成

    bandwidth = cluster.estimate_bandwidth(X, quantile=params[‘quantile’])

    # クラスタリングモデルの作成

    ms = cluster.MeanShift(bandwidth=bandwidth, bin_seeding=True)

    two_means = cluster.KMeans(n_clusters=params[‘n_clusters’])

    spectral = cluster.SpectralClustering(

        n_clusters=params[‘n_clusters’], eigen_solver=’arpack’,

        affinity=”nearest_neighbors”)

    gmm = mixture.GaussianMixture(

        n_components=params[‘n_clusters’], covariance_type=’full’)

    clustering_algorithms = (

        (‘KMeans’, two_means),

        (‘SpectralClustering’, spectral),

        (‘GaussianMixture’, gmm)

    )

    for name, algorithm in clustering_algorithms:

        t0 = time.time()

        # kneighbors_graphに関連する警告をキャッチ

        with warnings.catch_warnings():

            warnings.filterwarnings(

                “ignore”,

                message=”the number of connected components of the ” +

                “connectivity matrix is [0-9]{1,2}” +

                ” > 1. Completing it to avoid stopping the tree early.”,

                category=UserWarning)

            warnings.filterwarnings(

                “ignore”,

                message=”Graph is not fully connected, spectral embedding” +

                ” may not work as expected.”,

                category=UserWarning)

            algorithm.fit(X)

        t1 = time.time()

        if hasattr(algorithm, ‘labels_’):

            y_pred = algorithm.labels_.astype(np.int)

        else:

            y_pred = algorithm.predict(X)

# 結果の図を作成する

        plt.subplot(len(datasets), len(clustering_algorithms), plot_num)

        if i_dataset == 0:

            plt.title(name, size=30)

        colors = np.array(list(islice(cycle([‘#377eb8’, ‘#ff7f00’, ‘#4daf4a’,

                                             ‘#f781bf’, ‘#a65628’, ‘#984ea3’,

                                             ‘#999999’, ‘#e41a1c’, ‘#dede00’]),

                                      int(max(y_pred) + 1))))

        plt.scatter(X[:, 0], X[:, 1], s=10, color=colors[y_pred])

        plt.xlim(-2.5, 2.5)

        plt.ylim(-2.5, 2.5)

        plt.xticks(())

        plt.yticks(())

        plt.text(.99, .01, (‘%.2fs’ % (t1 – t0)).lstrip(‘0’),

                 transform=plt.gca().transAxes, size=30,

                 horizontalalignment=’right’)

        plot_num += 1

plt.show()