【天池入门笔记】【算法入门】sklearn入门系列二：聚类算法与特征选择

聚类算法主要有三种：层次聚类，划分聚类(sklearn)，密度聚类（DBSCAN）

1、聚类

#层次聚类
from sklearn.cluster import Agglomerative Clustering
import pandas as pd
from sklearn.preprocessing import StandardScaler
data = pd.read_csv('data.csv').fillna(0)
label = data.label
feature = data.drop('label',axis=1)
feature = StandardScaler().fit_transform(feature)
cluster = AgglomerativeClustering(n_clusters=2)
cluster.fit(feature)
pred = cluster.fit_predict(feature)

from sklearn.metrics import accuracy_score
print(accuracy_score(label,pred))

#划分聚类（kmeans）
cluster = KMeans(n_clusters=2,n_jobs=-1,init='k_means++')

#密度聚类（DBSCAN）
from sklearn import DBSCAN
cluster = DBSCAN(n_jobs=-1,eps=0.01)
pred = cluster.fit_predict(feature)

2、特征选择

#采用pearsonr相关系数选特征
import numpy as np
import pandas as pd
data.label.replace(-1,0,inplace = True)
data = data.fillna(0)
y = data.label
x = data.drop('label',axis=1)

from sklearn.model_selection import train_test_split
X_train,X_test,y_train,y_test = train_test_split（x,y,test_size=0.1,random_state=0）10%的数据作为测试集

from scipy.stats import pearsonr
columns = X_train.columns
pearsonr()
feature_importance = [(column,pearsonr(X_train[column],y_train)[0]) for column in columns]
#pearsonr()函数返回pearsonr值和p值，我们只需要pearsonr值，故取[0]

feature_importance.sort(key = lambda x:x[1])
#lambda函数是取其pearsonr值进行排序，丢弃column

 

#采用xgboost检验一下特征选择效果

import xgboost as xgb
dtrain = xbg.DMatrix(X_train,label=y_train)
dtest = xgb.DMatrix(X_test,label=y_test)

params = {
    'booster':'gbtree',
    'objective':'rank:pairwise',
    'eval_metric':'auc',
    'gamma':0.1,
    'min_child_weight':2,
    'max_depth':5,
    'lambda':10,
    'subsample':0.7,
    'colsample bytree':0.7,
    'eta':0.01,
    'tree_method':'exact',
    'seed':0,
    'nthead':7
}
watchlist = [(dtrain,'train'),(dtest,'test')]
model = xgb.train(params,dtrain,num_boost_round=100,evals=watchlist)


#再看一下删除相关系数小的特征之后的结果
#查看feature_importance，发现['merchant_max_distance']的pearsonr值较小
delete_feature = ['merchant_max_distance']
X_train = X_train[[for i in columns if i not in delete_feature]]
X_test = X_test[[for i in columns if i not in delete_feature]]

dtrain = xbg.DMatrix(X_train,label=y_train)
dtest = xgb.DMatrix(X_test,label=y_test)

watchlist = [(dtrain,'train'),(dtest,'test')]
model = xgb.train(params,dtrain,num_boost_round=100,evals=watchlist)

#运行之后比较出来的结果的auc值

 

#使用模型进行特征选择

#LogisticRegression
from sklearn.metrics import roc_auc_score
from sklearn.linear_model import LogisticRegression 
X_train,X_test,y_train,y_test = train_test_split(x,y,test_size=0.1,random_state=0)
lr = LogisticRegression(penalty='l2',random_state=0,n_jobs=-1).fit(X_train,y_train)
pred = lr.predict_proba(X_test)[:,1]
print(roc_auc_score(y_test,pred))

#Lasso
from sklearn.linear_model import RandomizedLasso 
from sklearn.datasets import load_boston
boston = load_boston()
X = boston['data']
Y = boston['target']
names = boston['feature_names']
rlasso = RandomizedLasso(alpha=0.025).fit(X,Y)
feature_importance = sorted(zip(names,rlasso.scores_))

#RFE
from sklearn.feature selection import RFE
X_train,X_test,y_train,y_test = train_test_split(x,y,test_size=0.1,random_state=0)
rf = RandomForestClassifier
rfe =fit(X_train,y_train)
feature_importance = sorted(zip(
    map(lambda x:round(x,4),rfe.ranking_),columns),reverse=True)

 

#SVC
cls = SVC(probability = True,kernel = 'rbf',c=0.1,max_iter=10)
cls.fit(X_train,y_train)
y_pred = cls.predict_proba(X_test)[:,1]
metrics.roc_auc_score(y_test)


#MLPRegresson
from sklearn.neural_network import MLPClassifier,MLPRegression
reg = MLPRegression(hidden_layer_sizes = (10,10,10),learning_rate = 0.1)


#DecisionTreeClassifier
from sklearn.tree import DecisionTreeClassifier
cls = DecisionTreeClassifier(max_depth=6,min_samples_split=10,
                             min_samples_leaf=5,max_features=0.7)
cls.fit(X_train,y_train)
y_pred = cls.predict_proba(X_test)[:,1]
metrics.roc_auc_score(y_test)


#RandomForestClassifier
cls = RandomForestClassifier(max_depth=6,min_samples_split=10,
                             min_samples_leaf=5,max_features=0.7)
cls.fit(X_train,y_train)
y_pred = cls.predict_proba(X_test)[:,1]
metrics.roc_auc_score(y_test)


#ExtraTreesClassifier