聚类算法

K-means算法

主要代码与接口

from sklearn.datasets import make_blobs
import matplotlib.pyplot as plt
from sklearn.cluster import KMeans
import pandas as pd
#X,y 数据
n_cluster = 4 #选择分类数目
#建模
cluster = KMeans(n_clusters = n_cluster,random_state = 0).fit(X)

#查看聚类结果
y_pred = cluster.labels_

#利用DataFrame查看结果分类情况
res = pd.DataFrame(y_pred)
res.iloc[:,0].unique()

#KMeans因为不需要建立模型或者预测结果，因此我们只要fit就能够得到聚类结果了
#KMeans也有接口predict和fit_predict，表示学习数据X并对X的类进行预测
#但所得到的结果和我们不调用predict,直接fit之后调用属性Labels一模一样
pre = cluster1.fit_predict(X)

#查看类的中心(聚类点坐标中心，也可以作为某一类的特征)
centroid = cluster1.cluster_centers_ 
centroid

标准score：轮廓系数

from sklearn.metrics import silhouette_score
from sklearn.metrics import silhouette_samples

silhouette_score(X,pre) #越接近1越好

自动化暴力调参

from sklearn.datasets import make_blobs
import matplotlib.pyplot as plt
from sklearn.cluster import KMeans
import pandas as pd
from sklearn.metrics import silhouette_score
from sklearn.metrics import silhouette_samples

maxScore = -float('inf')
for n_clusters in range(3,20):
    cluster = KMeans(n_clusters = n_clusters,random_state = 0).fit(X)
    pre_n = cluster.labels_
    curScore = silhouette_score(X,pre_n)
    if curScore > maxScore:
        maxScore = curScore
        max_clusters = n_clusters
print("最大轮廓系数：",maxScore,"最优簇数：",max_clusters)

DBSCAN算法

主要代码与接口

import matplotlib.pyplot as plt
from sklearn.cluster import DBSCAN
import pandas as pd
import numpy as np
from sklearn import metrics
import seaborn as sns
from sklearn.datasets.samples_generator import make_blobs
from sklearn.preprocessing import StandardScaler
#X,y为数据

#建模
db = DBSCAN(eps=i, min_samples=j).fit(X) #因为不同的参数组合，有可能导致计算得分出错，所以用try

#进行预测
pre_n= db.labels_#得到DBSCAN预测的分类便签

#评判标准即轮廓系数
curScore=metrics.silhouette_score(X_s,pre_n) #轮廓系数评价聚类的好坏，值越大越好

#计算噪声点个数占总数的比例
raito = len(pre_n[pre_n[:] == -1]) / len(labels)  #结果为-1的点为噪声点

# 获取分簇的数目
n_clusters_ = len(set(pre_n)) - (1 if -1 in pre_n else 0) 

print([i,j,k,raito,n_clusters_])

自动化暴力调参

import matplotlib.pyplot as plt
from sklearn.cluster import DBSCAN
import pandas as pd
import numpy as np
from sklearn import metrics
import seaborn as sns
from sklearn.datasets.samples_generator import make_blobs
from sklearn.preprocessing import StandardScaler

rs= []#存放各个参数的组合计算出来的模型评估得分和噪声比
eps = np.arange(5,50,1) #eps参数从0.2开始到4，每隔0.1进行一次
min_samples=np.arange(2,20,1)#min_samples参数从2开始到20

best_score=0
best_score_eps=0
best_score_min_samples=0

for i in eps:
     for j in min_samples:
         try:#因为不同的参数组合，有可能导致计算得分出错，所以用try
             db = DBSCAN(eps=i, min_samples=j).fit(X)
             pre_n= db.labels_#得到DBSCAN预测的分类便签
             curScore=metrics.silhouette_score(X_s,pre_n) #轮廓系数评价聚类的好坏，值越大越好
             raito = len(pre_n[pre_n[:] == -1]) / len(labels) #计算噪声点个数占总数的比例
             n_clusters_ = len(set(pre_n)) - (1 if -1 in pre_n else 0) # 获取分簇的数目
             rs.append([i,j,k,raito,n_clusters_])
            
             if curScore>best_score:
                            best_score=curScore
                            best_score_eps=i
                            best_score_min_samples=j

         except:
                    db='' #这里用try就是遍历i，j 计算轮廓系数会出错的，出错的就跳过
         else:
                    db=''
                
print(best_score
      ,best_score_eps
      ,best_score_min_samples)