带病在网吧里写，，，，给点鼓励吧。。。

http://scikit-learn.org/stable/modules/pipeline.html

1、pipeline和featureUnion是干什么的：

pipeline之前已经介绍过了，结合transformer和estimator。

featureUinon听名字就知道，将多个transformer的结果vector拼接成大的vector。

2、两者的区别：

前者相当于feature串行处理，后一个transformer处理前一个transformer的feature结果；

后者相当于feature的并行处理，将所有transformer的处理结果拼接成大的feature vector。

3、pipeline：chaining estimators

Pipelinecan be used to chain multiple estimators into one. 因为我们处理数据的过程一般都是比较固定的，比如特征选择、规范化、分类。所以pipeline主要由两个目的：

方便：fit、predict一次即可处理所有estimators的结果。

拼接参数选择：仅需一次即可grid search所有estimators的所有parameters。

pipeline的所有的estimators（除了最后一个）都必须是transformer（有transform方法），最后一个estimator可以使任何类型（transformer、classifier）

使用：通过一组（key,value）对来串联所有的estimators，key是自己对每一步骤的随意的命名，value是一个estimator object，例如：

>>> from sklearn.pipeline import Pipeline
sklearn.svm SVC
sklearn.decomposition PCA
>>> estimators = [('reduce_dim', PCA()), ('svm'SVC())]
clf = Pipeline(estimators)
clf 
Pipeline(steps=[('reduce_dim',PCA(copy=True,n_components=None,
 whiten=False)),('svm',SVC(C=1.0,cache_size=200,class_weight=None,
    coef0=0.0,degree=3,gamma=0.0,kernel='rbf',max_iter=-1,
    probability=False,random_state=None,shrinking=True,tol=0.001,51)">    verbose=False))])

>>> clf.steps[0]
('reduce_dim',whiten=False))

.named_steps['reduce_dim']
PCA(copy=True,whiten=False)

clf.set_paramssvm__C=10) 
    whiten=False)),SVC(C=10,51)">    verbose=False))])

终极目的，grid searches：

from sklearn.grid_search import GridSearchCV
>>> params = dict(reduce_dim__n_components=[2, 5,80)">10],
...               svm__C=[0.1,80)">10,80)">100])
>>> grid_search = GridSearchCV(clf, param_grid=params)

最经典的文本分类来了：

@H_301_349@# define a pipeline combining a text feature extractor with a simple
@H_301_349@# classifier
pipeline = Pipeline([
    ('vect', CountVectorizer()),
    ('tfidf', TfidfTransformer()),160)">'clf', SGDClassifier()),
])

@H_301_349@# uncommenting more parameters will give better exploring power but will
@H_301_349@# increase processing time in a combinatorial way
parameters = {
    'vect__max_df': (0.5, 0.75,80)">1.0),
    @H_301_349@#'vect__max_features': (None,5000,10000,50000),
    'vect__ngram_range': ((1,80)">1), (2)),  @H_301_349@# unigrams or bigrams
    @H_301_349@#'tfidf__use_idf': (True,False),144); font-style:italic">#'tfidf__norm': ('l1','l2'),160)">'clf__alpha': (0.00001,80)">0.000001),160)">'clf__penalty': ('l2', 'elasticnet'),144); font-style:italic">#'clf__n_iter': (10,50,80),
}

if __name__ == "__main__":
    @H_301_349@# multiprocessing requires the fork to happen in a __main__ protected
    @H_301_349@# block

    @H_301_349@# find the best parameters for both the feature extraction and the
    @H_301_349@# classifier
    grid_search = GridSearchCV(pipeline, parameters, n_jobs=-verbose1)

Notes：重要的事情不翻译，

Callingfiton the pipeline is the same as callingon each estimator in turn,transformthe input and pass it on to the next step.

sklearn.pipeline import FeatureUnion
sklearn.decomposition import PCA
import KernelPCA
>>> estimators = [('linear_pca', PCA()), ('kernel_pca', KernelPCA())]
>>> combined = FeatureUnion(estimators)
>>> combined 
FeatureUnion(n_jobs=1,transformer_list=[('linear_pca',
    n_components=None,whiten=False)),('kernel_pca',KernelPCA(alpha=1.0,51)">    coef0=1,eigen_solver='auto',fit_inverse_transform=False,51)">    gamma=None,kernel='linear',kernel_params=None,max_iter=None,remove_zero_eig=False,tol=0))],51)">    transformer_weights=None)

Author: Andreas Mueller <amueller@ais.uni-bonn.de>

# Author: Andreas Mueller <amueller@ais.uni-bonn.de>
@H_301_349@#
@H_301_349@# License: BSD 3 clause

import Pipeline, FeatureUnion
import GridSearchCV
sklearn.svm import SVC
sklearn.datasets import load_iris
import PCA
sklearn.feature_selection import SelectKBest

iris = load_iris()

X, y = iris.data, iris.target

@H_301_349@# This dataset is way to high-dimensional. Better do PCA:
pca = PCA(n_components2)

@H_301_349@# Maybe some original features where good,too?
selection = SelectKBest(k1)

@H_301_349@# Build estimator from PCA and Univariate selection:

combined_features = FeatureUnion([("pca", pca),160)">"univ_select", selection)])

@H_301_349@# Use combined features to transform dataset:
X_features = combined_features.fit(X, y).transform(X)

svm = SVC(kernel="linear")

@H_301_349@# Do grid search over k,n_components and C:

pipeline Pipeline([("features", combined_features),160)">"svm", svm)])

param_grid dict(features__pca__n_components3],
                  features__univ_select__k2],
                  svm__C10])

grid_search = GridSearchCV(pipeline,102)">=param_grid,80)">10)
grid_searchy)
print(grid_search.best_estimator_)