数据分析-流水线-pipeline_数据集划分 pipeline standardscaler

目录

前言

Pipeline

 网格搜索中使用流水线

流水线中模型选择

含特征选择的流水线

含PCA的流水线

组合流水线

---

前言

Pipeline

1. 数据预处理学习器 （如数据标准化、数据编码等，学习器必须有transform方法，用于数据转换）
2. 特征选择学习器 （学习器必须有transform方法，用于数据转换）
3. 执行预测的学习器 

以下流水线将标准化与KNN分类封装在一起。

from sklearn.pipeline import Pipeline
# 构建流水线
pipe = Pipeline(steps=[('scaler',StandardScaler()), 
                       ('knn', KNeighborsClassifier())])
# 训练
pipe.fit(X_train, y_train)
# 评估
print("测试集分类正确率：", round(pipe.score(X_test, y_test), 2))

 网格搜索中使用流水线

from sklearn.model_selection import GridSearchCV
# 设置参数网络，knn流水线中对KNN分类的命名knn__(双下划线)接对应模型的参数
param_grid = {'knn__n_neighbors': [2, 4, 6, 8, 10],
              'knn__weights': ['uniform', 'distance']}
# 网格搜索
grid_search = GridSearchCV(estimator=pipe, param_grid=param_grid, cv=5)  #cv=5 5折交叉验证
grid_search.fit(X_train, y_train)
# 测试集上的得分
grid_search.score(X_test, y_test)

含模型选择的流水线

实现

from sklearn.preprocessing import MinMaxScaler
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier
from sklearn.svm import SVC
from sklearn.pipeline import Pipeline
pipe=Pipeline(steps=[("scaler",MinMaxScaler()),("model",LogisticRegression())])
scale_selector=[StandardScaler(),MinMaxScaler()]
model_selector=[KNeighborsClassifier(),SVC(),LogisticRegression()]
param_grid={"scaler":scale_selector,"model":model_selector}
grid_search=GridSearchCV(estimator=pipe,param_grid=param_grid,cv=5)
grid_search.fit(X_train_s,y_train)
print(grid_search.best_estimator_)
grid_search.score(X_test_s,y_test)

  含特征选择的流水线

实现

from sklearn.feature_selection import RFECV
from sklearn.tree import DecisionTreeClassifier
 
# 在流水线中加入特征选择
pipe_new = Pipeline(steps=[('scaler',StandardScaler()), 
                           ('selector', RFECV(DecisionTreeClassifier(random_state=10), cv=5)),
                           ('model', KNeighborsClassifier())])
scale_selector=[StandardScaler(),MinMaxScaler()]
model_selector=[KNeighborsClassifier(),SVC(),LogisticRegression()]
# 设置参数网络
param_grid = {'scaler':scale_selector,
              'model': model_selector,
              'model__class_weight':['balanced', None],
              'model__C':[0.01, 0.1, 0.2, 0.5, 1]}  
# 网格搜索
grid_search = GridSearchCV(estimator=pipe_new, param_grid=param_grid, cv=5)
grid_search.fit(X_train, y_train)
 
# 输出最优的步骤，查看特征排名
pd.Series(grid_search.best_estimator_.named_steps['selector'].ranking_, index=X_train.columns)

含PCA的流水线

 实现

from sklearn.decomposition import PCA
 
# 在管道中加入PCA
pipe_new = Pipeline(steps=[('scaler',StandardScaler()), 
                           ('decomposition', PCA(3)),
                           ('model', KNeighborsClassifier())])
 
# 设置参数网络
param_grid = {'scaler':scale_selector,
              'model': model_selector,
              'decomposition__n_components':[2, 3, 4, 5, 6],
              'model__class_weight':['balanced', None],
              'model__C':[0.01, 0.1, 0.2, 0.5, 1]}  
# 网格搜索
grid_search = GridSearchCV(estimator=pipe_new, param_grid=param_grid, cv=5)
grid_search.fit(X_train, y_train)   
 
# 查看方差贡献率
grid_search.best_estimator_.named_steps['decomposition'].explained_variance_ratio_.sum()

复杂流水线

实现

from sklearn.decomposition import PCA
from sklearn.feature_selection import RFECV
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier
pipe_new2=Pipeline(steps=[("scaler",StandardScaler),("selector",PCA(3)),("model",KNeighborsClassifier())])
model_selector=[LogisticRegression(random_state=10),SVC(),KNeighborsClassifier()]
scaler_selector=[StandardScaler(),MinMaxScaler()]
selector_selector=[PCA(3),RFECV(DecisionTreeClassifier(random_state=10),cv=5)]
param_grid_2={"scaler":scaler_selector,"selector":selector_selector,"model":model_selector
             ,"model__class_weight":["balanced",None],
            "model__C":[0.01,0.1,0.2,0.5,1]}
grid_search=GridSearchCV(estimator=pipe_new2,param_grid=param_grid_2,cv=5)
grid_search.fit(X_train,y_train)
grid_search.best_estimator_.named_steps["selector"].explained_variance_ratio_.sum()