クラスター数の決め方の１つシルエット分析

# ライブラリ

from __future__ import print_function

from sklearn.datasets import make_blobs

from sklearn.cluster import KMeans

from sklearn.metrics import silhouette_samples, silhouette_score

import matplotlib.pyplot as plt

import matplotlib.cm as cm

import numpy as np

print(__doc__)

# サンプルの作成

X, y = make_blobs(n_samples=500,

                  n_features=2,

                  centers=4,

                  cluster_std=1,

                  center_box=(-10.0, 10.0),

                  shuffle=True,

                  random_state=1) 

range_n_clusters = [2, 3, 4, 5, 6]

for n_clusters in range_n_clusters:

    fig, (ax1, ax2) = plt.subplots(1, 2)

    fig.set_size_inches(18, 7)

    # シルエット計算と図

    ax1.set_xlim([-0.1, 1])

    ax1.set_ylim([0, len(X) + (n_clusters + 1) * 10])

    # K Meansクラスタs区政

    clusterer = KMeans(n_clusters=n_clusters, random_state=10)

    cluster_labels = clusterer.fit_predict(X)

    silhouette_avg = silhouette_score(X, cluster_labels)

    print(“For n_clusters =”, n_clusters,

          “The average silhouette_score is :”, silhouette_avg)

    #　シルエットスコア

    sample_silhouette_values = silhouette_samples(X, cluster_labels)

    y_lower = 10

    for i in range(n_clusters):

        ith_cluster_silhouette_values =

            sample_silhouette_values[cluster_labels == i]

        ith_cluster_silhouette_values.sort()

        size_cluster_i = ith_cluster_silhouette_values.shape[0]

        y_upper = y_lower + size_cluster_i

        color = cm.nipy_spectral(float(i) / n_clusters)

        ax1.fill_betweenx(np.arange(y_lower, y_upper),

                          0, ith_cluster_silhouette_values,

                          facecolor=color, edgecolor=color, alpha=0.7)

        ax1.text(-0.05, y_lower + 0.5 * size_cluster_i, str(i))

        y_lower = y_upper + 10 

# クラスタ図の作成

    ax1.set_title(“The silhouette plot for the various clusters.”)

    ax1.set_xlabel(“The silhouette coefficient values”)

    ax1.set_ylabel(“Cluster label”)

    ax1.axvline(x=silhouette_avg, color=”red”, linestyle=”–“)

    ax1.set_yticks([])  # Clear the yaxis labels / ticks

    ax1.set_xticks([-0.1, 0, 0.2, 0.4, 0.6, 0.8, 1])

    # 2nd Plot showing the actual clusters formed

    colors = cm.nipy_spectral(cluster_labels.astype(float) / n_clusters)

    ax2.scatter(X[:, 0], X[:, 1], marker=’.’, s=30, lw=0, alpha=0.7,

                c=colors, edgecolor=’k’)

    centers = clusterer.cluster_centers_

    ax2.scatter(centers[:, 0], centers[:, 1], marker=’o’,

                c=”white”, alpha=1, s=200, edgecolor=’k’)

    for i, c in enumerate(centers):

        ax2.scatter(c[0], c[1], marker=’$%d$’ % i, alpha=1,

                    s=50, edgecolor=’k’)

    ax2.set_title(“The visualization of the clustered data.”)

    ax2.set_xlabel(“Feature space for the 1st feature”)

    ax2.set_ylabel(“Feature space for the 2nd feature”)

    plt.suptitle((“Silhouette analysis for KMeans clustering on sample data “

                  “with n_clusters = %d” % n_clusters),

                 fontsize=14, fontweight=’bold’)

plt.show()