写在前面
scikit-learn 官网的Ensemble methods 文档部分只介绍了/bagging / boosting / voting / 三种模型组合方式;但是通过查找学习,受周志华《机器学习》集成学习部分的学习法启发,了解并学习了 stacking,在此以作记录。
概述
Stacking 是一种集合学习技术,通过元分类器组合多个分类模型。基于完整训练集训练各个分类模型; 然后,基于整体中的各个分类模型的输出 - 元特征来拟合元分类器。元分类器可以根据预测类标签或来自集合的概率进行训练。
流程图:
OR
[图片上传失败...(image-95bf0e-1518054917176)]
算法总结:
[图片上传失败...(image-f3b4e9-1518054917176)]
下面直接上实现过程
环境
- ubantu 16.04 + jupyter + python2.7
- scikit-learn + mlxtend + anconda
示例1.基础StackingClassifier
from sklearn import model_selection
from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier
from sklearn.naive_bayes import GaussianNB
from sklearn.ensemble import RandomForestClassifier
from mlxtend.classifier import StackingClassifier
import numpy as np
clf1 = KNeighborsClassifier(n_neighbors=1)
clf2 = RandomForestClassifier(random_state=1)
clf3 = GaussianNB()
lr = LogisticRegression()
sclf = StackingClassifier(classifiers=[clf1, clf2, clf3],
meta_classifier=lr)
print('3-fold cross validation:\n')
for clf, label in zip([clf1, clf2, clf3, sclf],
['KNN',
'Random Forest',
'Naive Bayes',
'StackingClassifier']):
scores = model_selection.cross_val_score(clf, X, y,
cv=3, scoring='accuracy')
print("Accuracy: %0.2f (+/- %0.2f) [%s]"
% (scores.mean(), scores.std(), label))
3-fold cross validation:
Accuracy: 0.91 (+/- 0.01) [KNN]
Accuracy: 0.91 (+/- 0.06) [Random Forest]
Accuracy: 0.92 (+/- 0.03) [Naive Bayes]
Accuracy: 0.95 (+/- 0.03) [StackingClassifier]
import matplotlib.pyplot as plt
from mlxtend.plotting import plot_decision_regions
import matplotlib.gridspec as gridspec
import itertools
gs = gridspec.GridSpec(2, 2)
fig = plt.figure(figsize=(10,8))
for clf, lab, grd in zip([clf1, clf2, clf3, sclf],
['KNN',
'Random Forest',
'Naive Bayes',
'StackingClassifier'],
itertools.product([0, 1], repeat=2)):
clf.fit(X, y)
ax = plt.subplot(gs[grd[0], grd[1]])
fig = plot_decision_regions(X=X, y=y, clf=clf)
plt.title(lab)
[图片上传失败...(image-2a710d-1518054917176)]
示例2.使用概率作为原特征的分类
或者,第一级分类器的类概率可用于通过设置来训练元分类器(第二级分类器)use_probas=True。如果average_probas=True,平均1级分类器的概率,如果average_probas=False,概率被学习法(推荐)。例如,在具有2个1级分类器的3类设置中,这些分类器可以对1个训练样本进行以下“概率”预测:
- 分类器1:[0.2,0.5,0.3]
- 分类器2:[0.3,0.4,0.4]
如果average_probas=True,元特征将是:
- [0.25,0.45,0.35]
相反,使用average_probas=Falsek个特征中的结果,其中,k = [n_classes * n_classifiers],通过学习法这些1级概率:
- [0.2,0.5,0.3,0.3,0.4,0.4]
clf1 = KNeighborsClassifier(n_neighbors=1)
clf2 = RandomForestClassifier(random_state=1)
clf3 = GaussianNB()
lr = LogisticRegression()
sclf = StackingClassifier(classifiers=[clf1, clf2, clf3],
use_probas=True,
average_probas=False,
meta_classifier=lr)
print('3-fold cross validation:\n')
for clf, label in zip([clf1, clf2, clf3, sclf],
['KNN',
'Random Forest',
'Naive Bayes',
'StackingClassifier']):
scores = model_selection.cross_val_score(clf, X, y,
cv=3, scoring='accuracy')
print("Accuracy: %0.2f (+/- %0.2f) [%s]"
% (scores.mean(), scores.std(), label))
3-fold cross validation:
Accuracy: 0.91 (+/- 0.01) [KNN]
Accuracy: 0.91 (+/- 0.06) [Random Forest]
Accuracy: 0.92 (+/- 0.03) [Naive Bayes]
Accuracy: 0.94 (+/- 0.03) [StackingClassifier]
示例3 - 学习法分类和GridSearch
要为scikit-learn设置参数网格GridSearch,需在参数网格中提供分类器的名称 - 在元回归的特殊情况下,添加'meta-'前缀。
from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier
from sklearn.naive_bayes import GaussianNB
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import GridSearchCV
from mlxtend.classifier import StackingClassifier
# Initializing models
clf1 = KNeighborsClassifier(n_neighbors=1)
clf2 = RandomForestClassifier(random_state=1)
clf3 = GaussianNB()
lr = LogisticRegression()
sclf = StackingClassifier(classifiers=[clf1, clf2, clf3],
meta_classifier=lr)
params = {'kneighborsclassifier__n_neighbors': [1, 5],
'randomforestclassifier__n_estimators': [10, 50],
'meta-logisticregression__C': [0.1, 10.0]}
grid = GridSearchCV(estimator=sclf,
param_grid=params,
cv=5,
refit=True)
grid.fit(X, y)
cv_keys = ('mean_test_score', 'std_test_score', 'params')
for r, _ in enumerate(grid.cv_results_['mean_test_score']):
print("%0.3f +/- %0.2f %r"
% (grid.cv_results_[cv_keys[0]][r],
grid.cv_results_[cv_keys[1]][r] / 2.0,
grid.cv_results_[cv_keys[2]][r]))
print('Best parameters: %s' % grid.best_params_)
print('Accuracy: %.2f' % grid.best_score_)
在此对于寻参方法,与之前的VoteClassifier 设计相同,但是作者本身还是更喜欢将一级分类器逐个寻优 ->之后才带入一级模型训练 - > 接着二级分类器模型寻参 -> 二级模型训练
如果我们计划多次使用回归算法,我们需要做的是在参数网格中添加一个额外的数字后缀,如下所示:
from sklearn.model_selection import GridSearchCV
# Initializing models
clf1 = KNeighborsClassifier(n_neighbors=1)
clf2 = RandomForestClassifier(random_state=1)
clf3 = GaussianNB()
lr = LogisticRegression()
sclf = StackingClassifier(classifiers=[clf1, clf1, clf2, clf3], # 此处变化
meta_classifier=lr)
params = {'kneighborsclassifier-1__n_neighbors': [1, 5],
'kneighborsclassifier-2__n_neighbors': [1, 5], # 此处变化
'randomforestclassifier__n_estimators': [10, 50],
'meta-logisticregression__C': [0.1, 10.0]}
grid = GridSearchCV(estimator=sclf,
param_grid=params,
cv=5,
refit=True)
grid.fit(X, y)
cv_keys = ('mean_test_score', 'std_test_score', 'params')
for r, _ in enumerate(grid.cv_results_['mean_test_score']):
print("%0.3f +/- %0.2f %r"
% (grid.cv_results_[cv_keys[0]][r],
grid.cv_results_[cv_keys[1]][r] / 2.0,
grid.cv_results_[cv_keys[2]][r]))
print('Best parameters: %s' % grid.best_params_)
print('Accuracy: %.2f' % grid.best_score_)
API 说明
StackingClassifier(classifiers,meta_classifier,use_probas = False,average_probas = False,verbose = 0)
参数
-
classifiers :array-like,shape = [n_classifiers]
一级分类器列表
-
meta_classifier :Object
二级分类器(元分类器)
-
use_probas :bool(默认值:False)
如果为True,则基于预测的概率而不是类标签来训练元分类器。
-
average_probas :bool(默认值:False)
如果为真,将概率平均为元特征。
-
verbose :int,optional(default = 0)
Controls the verbosity of the building process. - verbose=0 (default): Prints nothing - verbose=1: Prints the number & name of the regressor being fitted - verbose=2: Prints info about the parameters of the regressor being fitted - verbose>2: Changes verbose param of the underlying regressor to self.verbose - 2
属性
-
clfs_ :list,shape = [n_classifiers]
一级分类器
-
meta_clf_ :estimators器
二级分类器(元分类器)
方法
fit(X,y)
拟合合成分类器和元分类器。
Parameters
-
X :{array-like,sparse matrix},shape = [n_samples,n_features]
训练向量,其中n_samples是样本的数量,n_features是特征的数量。
-
y :array-like,shape = [n_samples]
target values。
Returns
self :Object
fit_transform(X,y = None,fit_params)
进行数据规整化
Parameters
-
X :numpy array of shape [n_samples, n_features]
训练集
-
y :numpy array of shape [n_samples]
标签
Returns
-
X_new :numpy array of shape [n_samples, n_features_new]
转换数组
get_params(deep = True)
返回GridSearch支持的estimators参数名称。
predict(X)
预测X的标签
Parameters
-
X :{array-like,sparse matrix},shape = [n_samples,n_features]
训练向量,其中n_samples是样本的数量,n_features是特征的数量。
Returns
- labels :array-like,shape = [n_samples]
Predicted class labels.
predict_proba(X)
Predict class probabilities for X.
Parameters
-
X :{array-like,sparse matrix},shape = [n_samples,n_features]
训练向量,其中n_samples是样本的数量,n_features是特征的数量。
Returns
-
proba :array-like,shape = [n_samples,n_classes]
每个样本的概率。
-
score(X,y,sample_weight = None)
Returns the mean accuracy on the given test data and labels.
在多标签分类中,这是子集精度,其是苛刻的度量,因为对于每个样本需要正确地预测每个标号集合。
Parameters
-
X :array-like,shape =(n_samples,n_features)
测试
-
y :array-like,shape =(n_samples)或(n_samples,n_outputs)
X的真实标签。
sample_weight :array-like,shape = [n_samples],可选
Returns
-
score :float
Mean accuracy of self.predict(X) wrt. y.
set_params(params)
设置estimators的参数。
该方法适用于简单estimators以及嵌套对象(例如pipelines)后者具有形式的参数