sklearn聚类+评价实例

模型

用了KMeans、DBSCAN、层次聚类法

KMeans模型

class sklearn.cluster.KMeans(n_clusters=8, init='k-means++', n_init=10, max_iter=300, tol=0.0001, precompute_distances='auto', verbose=0, random_state=None, copy_x=True, n_jobs=1)

重要的指标：n_clusters 聚成几类

属性与方法详解

DBSCAN模型

class sklearn.cluster.DBSCAN(eps=0.5, min_samples=5, metric='euclidean', algorithm='auto', leaf_size=30, p=None, random_state=None)

重要的指标：eps半径、min_samples半径之内的最小样本数

属性与方法详解

层次聚类模型

class sklearn.cluster.AgglomerativeClustering(n_clusters=2, affinity='euclidean', memory=Memory(cachedir=None), connectivity=None, n_components=None, compute_full_tree='auto', linkage='ward')

重要的指标：n_clusters 聚成几类

属性与方法详解

有自上而下分层次的，也有自下而上分层次的

sklearn中的聚类模型

sklearn.cluster类库

评价指标

sklearn.metrics类库

这里选用homogeneity_completeness_v_measure，输出
homogeneity、completeness、v_measure，v_measure是homogeneity、completeness的调和平均数

数据

import pandas as pd
import numpy as np
rawdata = pd.read_csv(r"..\Data\cluster_N.csv")
X = rawdata.iloc[:,0:-1]
Y = rawdata.iloc[:,-1]

没有缺失值和异常值，暂时不特征选择

这个数据是有真实标签的，依据真实标签对聚类效果的评价

建模

Kmeans建模+评价

from sklearn.cluster import KMeans, DBSCAN, AgglomerativeClustering
from sklearn.metrics import homogeneity_completeness_v_measure

def run_kmeans_model():
    model = KMeans(n_clusters=3).fit(X)
    result = homogeneity_completeness_v_measure(Y,model.labels_)[1:3]
    center = model.cluster_centers_
    return result, center

run_kmeans_model()

可视化

利用TSNE库将数据降维，并将聚类结果可视化

from sklearn.manifold import TSNE
import matplotlib.pyplot as plt

def draw(model):
    tsne = TSNE()
    tsne.fit_transform(X)
    tsne = pd.DataFrame(tsne.embedding_, index=X.index)
    r = tsne[model.labels_ == 0]
    plt.plot(r[0], r[1], "r.")
    g = tsne[model.labels_ == 1]
    plt.plot(g[0], g[1], "go")
    b = tsne[model.labels_ == 2]
    plt.plot(b[0], b[1], "b*")
    q = tsne[model.labels_ == -1]
    plt.plot(q[0], q[1], "cx")
    plt.show()

draw(KMeans(n_clusters=3).fit(X))

Kmeans结果

homogeneity = 0.7
completeness = 0.69

质心坐标

可视化结果

DBSCAN建模+评价

探究eps和n_samples对聚类的影响

def run_dbscan():
    result = {
        "min_samples":[],
        "eps":[],
        "n_clusters":[],
        "homogeneity":[],
        "completeness":[]
    }
    for i in range(5,11):
        for j in np.arange(0.5,1.1,0.1):
            model = DBSCAN(eps=j,min_samples=i,metric="euclidean").fit(X)  # 聚成5类
            homo, com = homogeneity_completeness_v_measure(Y, model.labels_)[1:3]
            n_cluster = pd.DataFrame(model.labels_).value_counts().shape[0]
            result["min_samples"].append(i)
            result["eps"].append(j)
            result["homogeneity"].append(homo)
            result["completeness"].append(com)
            result["n_clusters"].append(n_cluster)
    df = pd.DataFrame(result)
    return df

run_dbscan()

调优结果

目标是聚3簇，这里选择n_clusters要选4，因为DBSCAN有离群点

可视化

选择eps=0.9,min_samples=5建模可视化

此时 homogeneity = 0.51，completeness = 0.47

draw(DBSCAN(eps=0.9,min_samples=5,metric="euclidean").fit(X))

层次聚类建模+评价

def run_agg():
    model = AgglomerativeClustering(n_clusters=3).fit(X)
    result = homogeneity_completeness_v_measure(Y,model.labels_)[1:3]
    return result
    
run_agg()

homogeneity = 0.74
completeness = 0.73

可视化

draw(AgglomerativeClustering(n_clusters=3).fit(X))

# 系谱图
from scipy.cluster.hierarchy import linkage, dendrogram
from sklearn.preprocessing import MinMaxScaler

def drawc():
   df = MinMaxScaler().fit_transform(X)
   ss = linkage(df, method="ward")
   dendrogram(ss)
   plt.show()

drawc()

系谱图

结果

三个聚类模型的凝聚度和分离度