简介
Pipeline按顺序构建一系列转换和一个模型,最后的一步是模型。Pipeline中间的步骤必须是转换过程,它们必须包含fit和transform方法。最后一步模型只要有fit方法。
Pipeline的目的是能组合好几个步骤,当设置不同参数的时候,可以在一起做交叉验证。可以通过【pipeline的名称+ “__” + 参数名称】(注意是两个下划线)的方式设置多个步骤的参数。
参数
名称 | 类型 | 说明 |
---|---|---|
steps | list | 包含(name,transform)元组的list类型,按照元组的顺序形成一个链,最后一步是模型。 |
named_steps | dict | 只读的属性,用户通过设置的名称可以读取相应步骤的参数,keys是步骤名称,values是步骤参数 |
上手使用
from sklearn import svm
from sklearn.datasets import samples_generator
from sklearn.feature_selection import SelectKBest
from sklearn.feature_selection import f_regression
from sklearn.pipeline import Pipeline
from sklearn import svm
from sklearn.datasets import samples_generator
from sklearn.feature_selection import SelectKBest
from sklearn.feature_selection import f_regression
from sklearn.pipeline import Pipeline
产生一些测试数据
X, y = samples_generator.make_classification(n_informative=5, n_redundant=0, random_state=42)
X, y = samples_generator.make_classification(n_informative=5, n_redundant=0, random_state=42)
选择特征
# ANOVA SVM-C
anova_filter = SelectKBest(f_regression, k=5)
# ANOVA SVM-C
anova_filter = SelectKBest(f_regression, k=5)
SVM模型
clf = svm.SVC(kernel='linear')
clf = svm.SVC(kernel='linear')
构建pipeline
anova_svm = Pipeline([('anova', anova_filter), ('svc', clf)])
anova_svm = Pipeline([('anova', anova_filter), ('svc', clf)])
模型有两步,一步是最特征选择,一步是模型
设置参数
anova_svm.set_params(anova__k=10, svc__C=.1)
anova_svm.set_params(anova__k=10, svc__C=.1)
Pipeline(steps=[('anova', SelectKBest(k=10, score_func=<function f_regression at 0x4a0f0c8>)), ('svc', SVC(C=0.1, cache_size=200, class_weight=None, coef0=0.0,
decision_function_shape=None, degree=3, gamma='auto', kernel='linear',
max_iter=-1, probability=False, random_state=None, shrinking=True,
tol=0.001, verbose=False))])
训练模型
anova_svm.fit(X,y)
anova_svm.fit(X,y)
Pipeline(steps=[('anova', SelectKBest(k=10, score_func=<function f_regression at 0x4a0f0c8>)), ('svc', SVC(C=0.1, cache_size=200, class_weight=None, coef0=0.0,
decision_function_shape=None, degree=3, gamma='auto', kernel='linear',
max_iter=-1, probability=False, random_state=None, shrinking=True,
tol=0.001, verbose=False))])
预测结果
prediction = anova_svm.predict(X)
anova_svm.score(X,y)
0.77000000000000002
查看pipeline里的参数
anova_svm.named_steps['anova']
prediction = anova_svm.predict(X)
anova_svm.score(X,y)
0.77000000000000002
anova_svm.named_steps['anova']
SelectKBest(k=10, score_func=<function f_regression at 0x4a0f0c8>)
anova_svm.named_steps['svc']
SVC(C=0.1, cache_size=200, class_weight=None, coef0=0.0,
decision_function_shape=None, degree=3, gamma='auto', kernel='linear',
max_iter=-1, probability=False, random_state=None, shrinking=True,
tol=0.001, verbose=False)
anova_svm.named_steps['anova'].get_support()
array([ True, True, True, False, False, True, False, True, True,
True, False, False, True, False, True, False, False, False,
False, True], dtype=bool)
本文永久更新链接地址:http://www.linuxidc.com/Linux/2017-06/144939.htm