K-Means聚类分析-无标签

from sklearn.datasets import make_blobs
x,y=make_blobs(n_samples=600,n_features=10,centers=6,random_state=1) #10个特征值，可分为6簇

import pandas as pd
df_x=pd.DataFrame(x)
df_x.columns=['feature1','feature2','feature3','feature4','feature5','feature6','feature7','feature8','feature9','feature10']
df_y=pd.DataFrame(y)
df_y.columns=['class']
df_x.head()

from sklearn.model_selection import train_test_split
x_train,x_test,y_train,y_test=train_test_split(x,y,test_size=0.3,random_state=1)

from sklearn.cluster import KMeans
n=2 #初始参数设定
def kmeans(n,x_train,y_train):
    model=KMeans(n_clusters=n)
    model.fit(x_train,y_train)
    return model
model=kmeans(n,x_train,y_train)

from sklearn.metrics import silhouette_score
dic={} #定义空字典，存放轮廓系数
def lunkuo(model,x_test):
    y_pred=model.predict(x_test)
    print("轮廓系数：",round(silhouette_score(x_test,y_pred),2))
    dic[n]=(round(silhouette_score(x_test,y_pred),2))
    return model
print("-----初始聚成 2 簇-----")
model=lunkuo(model,x_test)

for n in range(3,9):
    print("-----聚成",n,"簇-----")
    model=kmeans(n,x_train,y_train) #训练模型
    model=lunkuo(model,x_test) #评估模型
print("-----最优参数-----")
print("最优簇数：",list(dic.keys())[list(dic.values()).index(max(dic.values()))]) #字典由value反查key
print("最优轮廓系数",max(dic.values())) #字典value中最大值

n=list(dic.keys())[list(dic.values()).index(max(dic.values()))] #最优簇数
model=kmeans(n,x_train,y_train) #训练最优模型
from sklearn.externals import joblib
joblib.dump(model,'d:\kmeans_nolabels.pkl')
new_model=joblib.load('d:\kmeans_nolabels.pkl')
new_model.predict(x_test)