top page> Aprendizaje automático> Agrupación> X-means

X-means

X-means es un tipo de algoritmo de agrupamiento que determina automáticamente el número de clústeres a medida que avanza el proceso de agrupamiento. Esta página compara los resultados de k-means++ y X-means.

Referencia de la clase pyclustering.cluster.xmeans.xmeans

Pelleg, Dan, y Andrew W. Moore. “X-means: Extending k-means with efficient estimation of the number of clusters.” Icml. Vol. 1. 2000.

import numpy as np
import matplotlib.pyplot as plt

from sklearn.cluster import KMeans
from sklearn.datasets import make_blobs

Cuando k está predefinido en k-means

def plot_by_kmeans(X, k=5):
    y_pred = KMeans(n_clusters=k, random_state=random_state, init="random").fit_predict(
        X
    )

    plt.scatter(X[:, 0], X[:, 1], c=y_pred, marker="x")
    plt.title(f"k-means, n_clusters={k}")


# Crear datos de muestra
n_samples = 1000
random_state = 117117
X, _ = make_blobs(
    n_samples=n_samples, random_state=random_state, cluster_std=1, centers=10
)

# Ejecutar k-means++.
plot_by_kmeans(X)

png

Ejecutar sin especificar el número de clústeres en X-means

BIC (criterio de información bayesiano)

from pyclustering.cluster.xmeans import xmeans
from pyclustering.cluster.center_initializer import kmeans_plusplus_initializer

BAYESIAN_INFORMATION_CRITERION = 0
MINIMUM_NOISELESS_DESCRIPTION_LENGTH = 1


def plot_by_xmeans(
    X, c_min=3, c_max=10, criterion=BAYESIAN_INFORMATION_CRITERION, tolerance=0.025
):
    initial_centers = kmeans_plusplus_initializer(X, c_min).initialize()
    xmeans_instance = xmeans(
        X, initial_centers, c_max, criterion=criterion, tolerance=tolerance
    )
    xmeans_instance.process()

    # Crear datos para las parcelas
    clusters = xmeans_instance.get_clusters()
    n_samples = X.shape[0]
    c = []
    for i, cluster_i in enumerate(clusters):
        X_ci = X[cluster_i]
        color_ci = [i for _ in cluster_i]
        plt.scatter(X_ci[:, 0], X_ci[:, 1], marker="x")
    plt.title("x-means")


# Ejecutar x-means
plot_by_xmeans(X, c_min=3, c_max=10, criterion=BAYESIAN_INFORMATION_CRITERION)

png

MINIMUM_NOISELESS_DESCRIPTION_LENGTH

plot_by_xmeans(X, c_min=3, c_max=10, criterion=MINIMUM_NOISELESS_DESCRIPTION_LENGTH)

png

Influencia del parámetro de tolerancia

X, _ = make_blobs(
    n_samples=2000,
    random_state=random_state,
    cluster_std=0.4,
    centers=10,
)

plt.figure(figsize=(25, 5))
for i, ti in enumerate(np.linspace(0.0001, 1, 5)):
    ti = np.round(ti, 4)
    plt.subplot(1, 10, i + 1)
    plot_by_xmeans(
        X, c_min=3, c_max=10, criterion=BAYESIAN_INFORMATION_CRITERION, tolerance=ti
    )
    plt.title(f"tol={ti}")

png

Comparar k-means y X-means para varios conjuntos de datos

for i in range(5):
    X, _ = make_blobs(
        n_samples=n_samples,
        random_state=random_state,
        cluster_std=0.7,
        centers=5 + i * 5,
    )
    plt.figure(figsize=(10, 5))
    plt.subplot(1, 2, 1)
    plot_by_kmeans(X)
    plt.subplot(1, 2, 2)
    plot_by_xmeans(X, c_min=3, c_max=20)
    plt.show()