[聚类算法]常用功能实现

前言：聚类是非监督学习的主要任务之一，根据原理可分为：基于质心、基于密度、基于连通性、基于概率以及基于神经网络等多种类型。

本文汇总了常用聚类算法及其评价指标，方便快速查询使用。（本文使用波士顿房价数据集，可用于回归）

以下为试验结果：

from time import time

import numpy as np
import pandas as pd
import matplotlib as mpl
import matplotlib.pyplot as plt
import sklearn
from sklearn import datasets

from sklearn.decomposition import PCA
from sklearn.preprocessing import scale

from sklearn import metrics
from sklearn.cluster import KMeans
from sklearn.cluster import MeanShift
from sklearn.cluster import DBSCAN
from sklearn.cluster import AgglomerativeClustering

#1. 加载数据
boston = sklearn.datasets.load_boston()
x,y = boston.data, boston.target
y = y.reshape(len(y),1)
data = np.hstack([x,y])

#2. 特征归一化
x = scale(x)
data = scale(data)


#3. 分析数据
name_data = boston.feature_names
#print(name_data)

df_x = pd.DataFrame(x,columns=name_data)
df_y = pd.DataFrame(y,columns=['MEDV'],dtype=np.int32)
df = pd.concat([df_x,df_y],axis=1)

# #506条数据，没有空值，float64类型
# print(df.head())
# print(df.info())
# print(df['MEDV'].describe())

#拟分为4类,目标为：<=17.025，>17.025 and <= 21.2, >21.2 and <=25, >25
n_clusters = 4#聚簇数量
df_y.loc[df_y['MEDV'] < 20] = 0
df_y.loc[(df_y['MEDV'] > 17.025) & (df_y['MEDV'] <= 21.2)] = 1
df_y.loc[(df_y['MEDV'] > 21.2) & (df_y['MEDV'] <= 25)] = 2
df_y.loc[df_y['MEDV'] > 25] = 3
labels = df_y.values.ravel()

def bench_k_means(estimator, name, data, method):
    t0 = time()
    estimator.fit(data)

    print('%-9s\t%-9s\t%.2fs\t\t%.3f\t\t\t%.3f\t\t%.3f\t\t%.3f\t\t\t%.3f\t\t\t%.3f'
          % (method, name, (time() - t0), 
             metrics.homogeneity_score(labels, estimator.labels_),
             metrics.completeness_score(labels, estimator.labels_),
             metrics.v_measure_score(labels, estimator.labels_),
             metrics.adjusted_rand_score(labels, estimator.labels_),
             metrics.adjusted_mutual_info_score(labels,  estimator.labels_,
                                                average_method='arithmetic'),
             metrics.silhouette_score(data, estimator.labels_,
                                      metric='euclidean',
                                      sample_size=300)))

print(115 * '_')
print('聚类方式\t\t聚类原理\t\t执行时间\t\t同质性得分\t\t完整性评分\tv-测量得分\t调整后兰德指数\t调整的相互信息\t轮廓系数')

# #5.1 KMeans
bench_k_means(KMeans(init='k-means++', n_clusters=n_clusters, n_init=10),
              name="质心", data=data, method='KMeans')

#5.2 KMeasn
bench_k_means(KMeans(init='random', n_clusters=n_clusters, n_init=10),
              name="质心", data=data, method='KMeans')

#5.3 KMeasn
pca = PCA(n_components=n_clusters).fit(data)
bench_k_means(KMeans(init=pca.components_, n_clusters=n_clusters, n_init=1),
              name="质心",
              data=data, method='KMeans')

#5.4 MeanShift
bench_k_means(MeanShift(),
              name="密度",
              data=data, method='MeanShift')

#5.5 DBSCAN
bench_k_means(DBSCAN(eps=3, min_samples=2),
              name="密度",
              data=data, method='DBSCAN')

#5.6 HCA
bench_k_means(AgglomerativeClustering(n_clusters=n_clusters),
              name="连通性",
              data=data, method='HCA')
print(115 * '_')