Skills of debugging model
Author: Xie Zhong-zhao
1. 特征提升
1.1 Feature Extraction
from sklearn.feature_extraction import DictVectorizer
'''
举个栗子，使用DictVectorizer对特征进行抽取和向量化
'''
measurements = [{'city':'Dubai','temperature':33.},{'city':'London','temperature':12.},{'city':'San Fransisco','temperature':18.}]
#初始化DictVectorizer特征抽取器
vec = DictVectorizer()
#输出转化之后的矩阵
print(vec.fit_transform(measurements).toarray())
#输出各个维度的特征含义
print(vec.get_feature_names())
'''
**************************************************************************
**************************************************************************
'''
'''
特征数值常见的计算方式有两种，分别是：CountVectorizer和TfidfVectorizer
(1)对每条训练文本，CountVectorizer只考虑每种词汇在该条训练文本中出现的频率
(2)TfidfVectorizer除了考量某个词汇在当前文本中出现的频率之外，同时关注包含这个词汇的文本条数的倒数。
'''
'''
使用CountVectorizer并且不去掉停用词的情况下，对文本特征进行量化的朴素贝叶斯分类性能测试
'''
#从sklearn.datasets里导入20类新闻文本数据抓取器
from sklearn.datasets import fetch_20newsgroups
#从互联网上及时下载新闻样本，subset='all'参数代表下载全部近2万条文本存储在变量news中
news = fetch_20newsgroups(subset='all')
#从sklearn.cross_validation导入train_test_split模块用于分割数据
from sklearn.cross_validation import train_test_split
#对news中数据data进行分割数据，25%的文本用作测试集；75%作为训练集
X_train,X_test,y_train,y_test = train_test_split(news.data, news.target,test_size=0.25,random_state=33)
#从sklearn.feature_extraction.text里导入CountVectorizer
from sklearn.feature_extraction.text import CountVectorizer
#采用默认设置对CountVectorizer进行初始化(默认配置不去除英文停用词)，并且赋值给变量count_vec
count_vec = CountVectorizer()
#只使用词频统计方式将原始训练和测试文本转化为特征向量
X_count_train = count_vec.fit_transform(X_train)
X_count_test = count_vec.transform(X_test)
#从sklearn.naive_bayes里导入朴素贝叶斯分类器
from sklearn.naive_bayes import MultinomialNB
#使用默认配置对分类器进行初始化
mnb_count = MultinomialNB()
#使用朴素贝叶斯分类器，对CountVectorizer(不去停用词)训练样本进行参数学习
mnb_count.fit(X_count_train,y_train)
#输出模型准确性结果
print('the accuracy of classifying 20newgroups using Naive Bayes(CountVectorizer without filtering stopwords):',
mnb_count.score(X_count_test,y_test))
#将分类预测结果存储在变量y_count_predict中
y_count_predict = mnb_count.predict(X_count_test)
#输出更加详细的其他评价的分类性能指标
from sklearn.metrics import classification_report
print(classification_report(y_test,y_count_predict,target_names=news.target_names))
'''
**************************************************************************
**************************************************************************
'''
'''
使用TfidfVectorizer并且不去掉停用词的情况下，对文本特征进行量化的朴素贝叶斯分类性能测试
'''
#从sklearn.feature_extraction.text里面分别导入TfidfVectorizer
from sklearn.feature_extraction.text import TfidfVectorizer
#采用默认配置对TfidfVectorizer进行初始化，并且赋值给变量tfidf_vec
tfidf_vec = TfidfVectorizer()
#使用tfidf的方式，将原始训练和测试的文本转化为特征向量
X_tfidf_train = tfidf_vec.fit_transform(X_train)
X_tfidf_test = tfidf_vec.transform(X_test)
#依然使用默认配置的朴素贝叶斯分类器，在相同的训练和测试数据上，对新的特征量化方式进行性能评估
mnb_tfidf = MultinomialNB()
mnb_tfidf.fit(X_tfidf_train,y_train)
#输出模型准确性结果
print('the accuracy of classifying 20newgroups using Naive Bayes(TfidfVectorizer without filtering stopwords):',
mnb_tfidf.score(X_tfidf_test,y_test))
y_tfidf_predict = mnb_tfidf.predict(X_tfidf_test)
#输出更加详细的其他评价的分类性能指标
print(classification_report(y_test,y_tfidf_predict,target_names=news.target_names))
'''
**************************************************************************
**************************************************************************
'''
'''
分别使用CountVectorizer与TfidfVectorizer,并且去掉停用词的条件下，对文本特征进行量化的朴素贝叶斯分类性能测试
'''
count_filter_vec,tfidf_filter_vec = CountVectorizer(analyzer='word',stop_words='english'),TfidfVectorizer(analyzer='word',stop_words='english')
#使用带有停用词过滤的CountVectorizer对训练和测试文本分别进行量化处理
X_count_filter_train = count_filter_vec.fit_transform(X_train)
X_count_filter_test = count_filter_vec.transform(X_test)
#使用带有停用词过滤的TfidfVectorizer对训练和测试样本分别进行量化处理
X_tfidf_filter_train = tfidf_filter_vec.fit_transform(X_train)
X_tfidf_filter_test = tfidf_filter_vec.transform(X_test)
#初始化默认配置朴素贝叶斯分类器，并对CountVectorizer后的数据进行预测与准确的评估
mnb_count_filter = MultinomialNB()
mnb_count_filter.fit(X_count_filter_train,y_train)
print('the accuracy of classifying 20newgroups using Naive Bayes(CountVectorizer by filtering stopwords):',
mnb_count_filter.score(X_count_filter_test,y_test))
y_count_filter_predict = mnb_count_filter.predict(X_count_filter_test)
print(classification_report(y_test,y_count_filter_predict,target_names=news.target_names))
#初始化默认配置朴素贝叶斯分类器，并对TfidfVectorizer后的数据进行预测与准确的评估
mnb_tfidf_filter = MultinomialNB()
mnb_tfidf_filter.fit(X_tfidf_filter_train,y_train)
print('the accuracy of classifying 20newgroups using Naive Bayes(tfidfVectorizer by filtering stopwords):',
mnb_tfidf_filter.score(X_tfidf_filter_test,y_test))
y_tfidf_filter_predict = mnb_tfidf_filter.predict(X_tfidf_filter_test)
print(classification_report(y_test,y_tfidf_filter_predict,target_names=news.target_names))[[ 1. 0. 0. 33.]
[ 0. 1. 0. 12.]
[ 0. 0. 1. 18.]]
['city=Dubai', 'city=London', 'city=San Fransisco', 'temperature']
the accuracy of classifying 20newgroups using Naive Bayes(CountVectorizer without filtering stopwords): 0.839770797963
precision recall f1-score support
alt.atheism 0.86 0.86 0.86 201
comp.graphics 0.59 0.86 0.70 250
comp.os.ms-windows.misc 0.89 0.10 0.17 248
comp.sys.ibm.pc.hardware 0.60 0.88 0.72 240
comp.sys.mac.hardware 0.93 0.78 0.85 242
comp.windows.x 0.82 0.84 0.83 263
misc.forsale 0.91 0.70 0.79 257
rec.autos 0.89 0.89 0.89 238
rec.motorcycles 0.98 0.92 0.95 276
rec.sport.baseball 0.98 0.91 0.95 251
rec.sport.hockey 0.93 0.99 0.96 233
sci.crypt 0.86 0.98 0.91 238
sci.electronics 0.85 0.88 0.86 249
sci.med 0.92 0.94 0.93 245
sci.space 0.89 0.96 0.92 221
soc.religion.christian 0.78 0.96 0.86 232
talk.politics.guns 0.88 0.96 0.92 251
talk.politics.mideast 0.90 0.98 0.94 231
talk.politics.misc 0.79 0.89 0.84 188
talk.religion.misc 0.93 0.44 0.60 158
avg / total 0.86 0.84 0.82 4712
the accuracy of classifying 20newgroups using Naive Bayes(TfidfVectorizer without filtering stopwords): 0.846349745331
precision recall f1-score support
alt.atheism 0.84 0.67 0.75 201
comp.graphics 0.85 0.74 0.79 250
comp.os.ms-windows.misc 0.82 0.85 0.83 248
comp.sys.ibm.pc.hardware 0.76 0.88 0.82 240
comp.sys.mac.hardware 0.94 0.84 0.89 242
comp.windows.x 0.96 0.84 0.89 263
misc.forsale 0.93 0.69 0.79 257
rec.autos 0.84 0.92 0.88 238
rec.motorcycles 0.98 0.92 0.95 276
rec.sport.baseball 0.96 0.91 0.94 251
rec.sport.hockey 0.88 0.99 0.93 233
sci.crypt 0.73 0.98 0.83 238
sci.electronics 0.91 0.83 0.87 249
sci.med 0.97 0.92 0.95 245
sci.space 0.89 0.96 0.93 221
soc.religion.christian 0.51 0.97 0.67 232
talk.politics.guns 0.83 0.96 0.89 251
talk.politics.mideast 0.92 0.97 0.95 231
talk.politics.misc 0.98 0.62 0.76 188
talk.religion.misc 0.93 0.16 0.28 158
avg / total 0.87 0.85 0.84 4712
the accuracy of classifying 20newgroups using Naive Bayes(CountVectorizer by filtering stopwords): 0.863752122241
precision recall f1-score support
alt.atheism 0.85 0.89 0.87 201
comp.graphics 0.62 0.88 0.73 250
comp.os.ms-windows.misc 0.93 0.22 0.36 248
comp.sys.ibm.pc.hardware 0.62 0.88 0.73 240
comp.sys.mac.hardware 0.93 0.85 0.89 242
comp.windows.x 0.82 0.85 0.84 263
misc.forsale 0.90 0.79 0.84 257
rec.autos 0.91 0.91 0.91 238
rec.motorcycles 0.98 0.94 0.96 276
rec.sport.baseball 0.98 0.92 0.95 251
rec.sport.hockey 0.92 0.99 0.95 233
sci.crypt 0.91 0.97 0.93 238
sci.electronics 0.87 0.89 0.88 249
sci.med 0.94 0.95 0.95 245
sci.space 0.91 0.96 0.93 221
soc.religion.christian 0.87 0.94 0.90 232
talk.politics.guns 0.89 0.96 0.93 251
talk.politics.mideast 0.95 0.98 0.97 231
talk.politics.misc 0.84 0.90 0.87 188
talk.religion.misc 0.91 0.53 0.67 158
avg / total 0.88 0.86 0.85 4712
the accuracy of classifying 20newgroups using Naive Bayes(tfidfVectorizer by filtering stopwords): 0.882640067912
precision recall f1-score support
alt.atheism 0.86 0.81 0.83 201
comp.graphics 0.85 0.81 0.83 250
comp.os.ms-windows.misc 0.84 0.87 0.86 248
comp.sys.ibm.pc.hardware 0.78 0.88 0.83 240
comp.sys.mac.hardware 0.92 0.90 0.91 242
comp.windows.x 0.95 0.88 0.91 263
misc.forsale 0.90 0.80 0.85 257
rec.autos 0.89 0.92 0.90 238
rec.motorcycles 0.98 0.94 0.96 276
rec.sport.baseball 0.97 0.93 0.95 251
rec.sport.hockey 0.88 0.99 0.93 233
sci.crypt 0.85 0.98 0.91 238
sci.electronics 0.93 0.86 0.89 249
sci.med 0.96 0.93 0.95 245
sci.space 0.90 0.97 0.93 221
soc.religion.christian 0.70 0.96 0.81 232
talk.politics.guns 0.84 0.98 0.90 251
talk.politics.mideast 0.92 0.99 0.95 231
talk.politics.misc 0.97 0.74 0.84 188
talk.religion.misc 0.96 0.29 0.45 158
avg / total 0.89 0.88 0.88 4712
1.2 Feature Selection
import pandas as pd
'''
使用Titanic数据集，通过特征筛选的方式一步步提升决策树的预测性能
'''
#从互联网读取titanic数据
titanic = pd.read_csv('http://biostat.mc.vanderbilt.edu/wiki/pub/Main/DataSets/titanic.txt')
#分离数据特征与预测目标
y = titanic['survived']
X = titanic.drop(['row.names','name','survived'],axis=1)
#对缺失的数据进行填充
X['age'].fillna(X['age'].mean(),inplace=True)
X.fillna('UNKNOW',inplace=True)
print(X.shape)
print(y.shape)
#分割数据，依然采用25%用于测试
from sklearn.cross_validation import train_test_split
X_train,X_test,y_train,y_test = train_test_split(X,y,test_size=0.25,random_state=33)
print(X_train)
print(X_train.shape)
#类别型特征向量化
from sklearn.feature_extraction import DictVectorizer
vec = DictVectorizer()
X_train = vec.fit_transform(X_train.to_dict(orient='record'))
X_test = vec.transform(X_test.to_dict(orient ='record'))
print(X_train)
print(X_train.shape)
print(len(vec.feature_names_))
# print(X_test)
# #输出处理后特征向量的维度
# print(len(vec.feature_names_))
#
#使用决策树模型依靠所有特征进行预测，并作性能评估
from sklearn.tree import DecisionTreeClassifier
dt = DecisionTreeClassifier(criterion='entropy')
dt.fit(X_train,y_train)
print('the accuracy of Decision Tree',dt.score(X_test,y_test))
#
#从sklearn导入特征筛选器
from sklearn import feature_selection
#筛选前20%的特征，使用相同的配置的决策树模型进行预测，并且评估性能
fs = feature_selection.SelectPercentile(feature_selection.chi2,percentile=20)
X_train_fs = fs.fit_transform(X_train,y_train)
dt.fit(X_train_fs,y_train)
X_test_fs = fs.transform(X_test)
print('the accuracy of selecting 20% features in model',dt.score(X_test_fs,y_test))
#
#通过交叉验证的方法，按照固定的百分比筛选特征，并作图展示性能随特征筛选比例的变化
from sklearn.cross_validation import cross_val_score
import numpy as np
percentiles = range(1,100,2)
results = []
for i in percentiles:
fs = feature_selection.SelectPercentile(feature_selection.chi2,percentile=i)
X_train_fs = fs.fit_transform(X_train,y_train)
scores = cross_val_score(dt,X_train_fs,y_train,cv=5)
results = np.append(results,scores.mean()) #列表拼接
print(results)
#找到提现最佳性能的特征筛选的百分比
# opt = np.where(results == results.max())[0]
# print('Optimal number of features %d'%percentiles[opt])
#
import pylab as pl
fig = pl.figure()
pl.plot(percentiles,results)
pl.xlabel('percentiles of features')
pl.ylabel('accuracy')
fig.savefig("feature selecting.png", dpi=300) #save graph
pl.show()
#使用最佳筛选后的特征，利用相同配置在测试集上进行性能和评估
from sklearn import feature_selection
fs = feature_selection.SelectPercentile(feature_selection.chi2,percentile=7)
X_train_fs = fs.fit_transform(X_train,y_train)
dt.fit(X_train_fs,y_train)
X_test_fs = fs.transform(X_test)
print(dt.score(X_test_fs,y_test))(1313, 8)
(1313,)
pclass age embarked home.dest \
1086 3rd 31.194181 UNKNOW UNKNOW
12 1st 31.194181 Cherbourg Paris, France
1036 3rd 31.194181 UNKNOW UNKNOW
833 3rd 32.000000 Southampton Foresvik, Norway Portland, ND
1108 3rd 31.194181 UNKNOW UNKNOW
562 2nd 41.000000 Cherbourg New York, NY
437 2nd 48.000000 Southampton Somerset / Bernardsville, NJ
663 3rd 26.000000 Southampton UNKNOW
669 3rd 19.000000 Southampton England
507 2nd 31.194181 Southampton Petworth, Sussex
1167 3rd 31.194181 UNKNOW UNKNOW
821 3rd 9.000000 Southampton Strood, Kent, England Detroit, MI
327 2nd 32.000000 Southampton Warwick, England
715 3rd 21.000000 Southampton Ireland New York, NY
308 1st 31.194181 UNKNOW UNKNOW
1274 3rd 31.194181 UNKNOW UNKNOW
640 3rd 40.000000 Southampton Sweden Worcester, MA
72 1st 70.000000 Southampton Milwaukee, WI
1268 3rd 31.194181 UNKNOW UNKNOW
1024 3rd 31.194181 UNKNOW UNKNOW
1047 3rd 31.194181 UNKNOW UNKNOW
940 3rd 31.194181 UNKNOW UNKNOW
350 2nd 20.000000 Southampton Skara, Sweden / Rockford, IL
892 3rd 31.194181 UNKNOW UNKNOW
555 2nd 30.000000 Southampton Finland / Washington, DC
176 1st 36.000000 Southampton Philadelphia, PA
107 1st 31.194181 Cherbourg New York, NY
475 2nd 34.000000 Southampton Chicago, IL
330 2nd 23.000000 Southampton Guernsey
533 2nd 34.000000 Southampton Denmark / New York, NY
... ... ... ... ...
235 1st 24.000000 Southampton Huntington, WV
465 2nd 22.000000 Southampton India / Pittsburgh, PA
210 1st 31.194181 Southampton New York, NY
579 2nd 40.000000 Southampton Aberdeen / Portland, OR
650 3rd 23.000000 Southampton UNKNOW
1031 3rd 31.194181 UNKNOW UNKNOW
99 1st 24.000000 Southampton Winnipeg, MB
969 3rd 31.194181 UNKNOW UNKNOW
535 2nd 31.194181 Cherbourg Paris
403 2nd 31.194181 Southampton England / Philadelphia, PA
744 3rd 45.000000 Southampton Australia Fingal, ND
344 2nd 26.000000 Southampton Elmira, NY / Orange, NJ
84 1st 31.194181 Southampton San Francisco, CA
528 2nd 20.000000 Southampton Gunnislake, England / Butte, MT
1270 3rd 31.194181 UNKNOW UNKNOW
662 3rd 40.000000 Cherbourg UNKNOW
395 2nd 42.000000 Southampton Greenport, NY
1196 3rd 31.194181 UNKNOW UNKNOW
543 2nd 23.000000 Cherbourg Paris / Montreal, PQ
845 3rd 31.194181 UNKNOW UNKNOW
813 3rd 25.000000 Queenstown New York, NY
61 1st 31.194181 Southampton St Leonards-on-Sea, England Ohio
102 1st 23.000000 Southampton Winnipeg, MB
195 1st 28.000000 UNKNOW ?Havana, Cuba
57 1st 27.000000 Southampton New York, NY / Ithaca, NY
1225 3rd 31.194181 UNKNOW UNKNOW
658 3rd 31.194181 Cherbourg Syria New York, NY
578 2nd 12.000000 Southampton Aberdeen / Portland, OR
391 2nd 18.000000 Southampton New Forest, England
1044 3rd 31.194181 UNKNOW UNKNOW
room ticket boat sex
1086 UNKNOW UNKNOW UNKNOW male
12 B-35 17477 L69 6s 9 female
1036 UNKNOW UNKNOW UNKNOW male
833 UNKNOW UNKNOW UNKNOW male
1108 UNKNOW UNKNOW UNKNOW male
562 UNKNOW UNKNOW UNKNOW male
437 UNKNOW UNKNOW 9 female
663 UNKNOW UNKNOW UNKNOW male
669 UNKNOW UNKNOW UNKNOW male
507 UNKNOW UNKNOW UNKNOW male
1167 UNKNOW UNKNOW UNKNOW male
821 UNKNOW UNKNOW C male
327 UNKNOW UNKNOW UNKNOW female
715 UNKNOW UNKNOW UNKNOW male
308 UNKNOW UNKNOW UNKNOW female
1274 UNKNOW UNKNOW UNKNOW male
640 UNKNOW UNKNOW (142) male
72 UNKNOW UNKNOW (269) male
1268 UNKNOW UNKNOW UNKNOW male
1024 UNKNOW UNKNOW UNKNOW male
1047 UNKNOW UNKNOW UNKNOW female
940 UNKNOW UNKNOW UNKNOW male
350 UNKNOW UNKNOW 12 female
892 UNKNOW UNKNOW UNKNOW male
555 UNKNOW UNKNOW UNKNOW female
176 UNKNOW UNKNOW 7 male
107 UNKNOW UNKNOW 5 female
475 F-33 UNKNOW 14/D female
330 UNKNOW UNKNOW UNKNOW male
533 UNKNOW 250647 UNKNOW male
... ... ... ... ...
235 UNKNOW UNKNOW UNKNOW male
465 UNKNOW UNKNOW UNKNOW female
210 UNKNOW UNKNOW 15 male
579 UNKNOW UNKNOW 9 female
650 UNKNOW UNKNOW UNKNOW male
1031 UNKNOW UNKNOW UNKNOW male
99 UNKNOW UNKNOW 10 female
969 UNKNOW UNKNOW UNKNOW male
535 UNKNOW UNKNOW UNKNOW male
403 UNKNOW UNKNOW (286) male
744 UNKNOW UNKNOW 15 male
344 UNKNOW UNKNOW UNKNOW male
84 UNKNOW UNKNOW 13 male
528 UNKNOW UNKNOW UNKNOW male
1270 UNKNOW UNKNOW UNKNOW male
662 UNKNOW UNKNOW UNKNOW male
395 UNKNOW 28220 L32 10s UNKNOW male
1196 UNKNOW UNKNOW UNKNOW male
543 UNKNOW UNKNOW UNKNOW male
845 UNKNOW UNKNOW UNKNOW male
813 UNKNOW UNKNOW UNKNOW male
61 UNKNOW UNKNOW 6 female
102 UNKNOW UNKNOW 10 female
195 UNKNOW UNKNOW (189) male
57 UNKNOW UNKNOW 5 male
1225 UNKNOW UNKNOW UNKNOW male
658 UNKNOW UNKNOW UNKNOW female
578 UNKNOW UNKNOW 9 female
391 UNKNOW UNKNOW UNKNOW male
1044 UNKNOW UNKNOW UNKNOW female
[984 rows x 8 columns]
(984, 8)
(0, 0) 31.1941810427
(0, 78) 1.0
(0, 82) 1.0
(0, 366) 1.0
(0, 391) 1.0
(0, 435) 1.0
(0, 437) 1.0
(0, 473) 1.0
(1, 0) 31.1941810427
(1, 73) 1.0
(1, 79) 1.0
(1, 296) 1.0
(1, 389) 1.0
(1, 397) 1.0
(1, 436) 1.0
(1, 446) 1.0
(2, 0) 31.1941810427
(2, 78) 1.0
(2, 82) 1.0
(2, 366) 1.0
(2, 391) 1.0
(2, 435) 1.0
(2, 437) 1.0
(2, 473) 1.0
(3, 0) 32.0
: :
(980, 473) 1.0
(981, 0) 12.0
(981, 73) 1.0
(981, 81) 1.0
(981, 84) 1.0
(981, 390) 1.0
(981, 435) 1.0
(981, 436) 1.0
(981, 473) 1.0
(982, 0) 18.0
(982, 78) 1.0
(982, 81) 1.0
(982, 277) 1.0
(982, 390) 1.0
(982, 435) 1.0
(982, 437) 1.0
(982, 473) 1.0
(983, 0) 31.1941810427
(983, 78) 1.0
(983, 82) 1.0
(983, 366) 1.0
(983, 391) 1.0
(983, 435) 1.0
(983, 436) 1.0
(983, 473) 1.0
(984, 474)
474
the accuracy of Decision Tree 0.80547112462
the accuracy of selecting 20% features in model 0.826747720365
[ 0.85063904 0.85673057 0.87602556 0.88622964 0.86082251 0.87302618
0.87100598 0.87302618 0.86693465 0.86894455 0.86895485 0.86491445
0.86387343 0.86793445 0.86082251 0.86486291 0.86283241 0.8628221
0.86283241 0.86083282 0.86690373 0.86997526 0.86693465 0.86793445
0.86792414 0.86892393 0.87200577 0.87097506 0.86691404 0.86284271
0.86794475 0.86997526 0.86998557 0.86996496 0.87098536 0.86487322
0.86692435 0.86794475 0.86895485 0.86588332 0.86588332 0.86791383
0.86692435 0.85776129 0.86692435 0.8618223 0.86286333 0.86082251
0.85982272 0.86386312]
0.857142857143
2. 模型正则化
2.1 Underfitting and Overfitting
#使用线性回归模型在比萨训练样本上进行拟合
from sklearn.linear_model import LinearRegression
'''
使用线性回归模型在比萨训练样本上进行拟合
'''
#输入训练样本的特征及其目标值，分别存储在变量X_train与y_train
X_train = [[6],[8],[10],[14],[18]]
y_train = [[7],[9],[13],[17.5],[18]]
#使用默认配置初始化线性回归模型
regressor = LinearRegression()
#直接以比萨的半径作为特征训练模型
regressor.fit(X_train,y_train)
import numpy as np
#在x轴上从0到25均匀采样100个数据点
xx = np.linspace(0,26,100)
# print(xx.shape[0])
xx = xx.reshape(xx.shape[0],1)
print(xx)
#以上述100个数据点作为基准，预测回归直线
yy = regressor.predict(xx)
#对回归预测到直线进行作图
import matplotlib.pyplot as plt
plt.scatter(X_train,y_train) #画出样本散点图
plt1,=plt.plot(xx,yy,label="Degree=1") #画出回归预测直线
plt.axis([0,25,0,25]) #设置坐标轴的范围
plt.xlabel('Diameter of Pizza')
plt.ylabel('Price of pizza')
plt.legend(handles=[plt1])
plt.show()
'''
输出在线性回归模型在训练样本上的R-squared值
'''
print('the R-squared value of Linear Regressor performing on the training data is:', regressor.score(X_train,y_train))
'''
*****************************************************************************************************
*****************************************************************************************************
*****************************************************************************************************
'''
'''
使用2次多项式回归模型在比萨训练样本上进行拟合
'''
#从sklearn.preprocessing中导入多项式特征产生器
from sklearn.preprocessing import PolynomialFeatures
#使用PolynomialFeatures(degree=2)映射出2次多项式特征，存储变量X_train_poly2中
poly2 = PolynomialFeatures(degree=2)
X_train_poly2 = poly2.fit_transform(X_train)
#以线性回归器为基础，初始化回归模型，尽管特征的维度有所提升，但是模型基础仍然是线性模型
regressor_poly2 = LinearRegression()
#对2次多项式回归模型进行训练
regressor_poly2.fit(X_train_poly2,y_train)
#从新映射绘图用x轴采样数据
xx_poly2 = poly2.transform(xx)
#使用2次多项式回归模型对应x轴采样数据进行回归预测
yy_poly2 = regressor_poly2.predict(xx_poly2)
'''
分别对训练数据点、线性回归直线、2次多项式回归曲线进行作图
'''
plt.scatter(X_train,y_train)
plt1,=plt.plot(xx,yy,label='Degree=1')
plt2,=plt.plot(xx,yy_poly2,label='Degree=2')
plt.axis([0,25,0,25])
plt.xlabel('Diameter of Pizza')
plt.ylabel('Price of Pizza')
plt.legend(handles=[plt1,plt2])
plt.show()
'''
输出2次多项式回归模型在训练样本上的R-squared值
'''
print('the R-squared value of Polynomial Regressor(Degree=2) performing on training data is:',
regressor_poly2.score(X_train_poly2,y_train))
'''
*****************************************************************************************************
*****************************************************************************************************
*****************************************************************************************************
'''
'''
使用4次多项式回归模型在披萨训练样本上进行拟合
'''
from sklearn.preprocessing import PolynomialFeatures
#使用PolynomialFratures将x特征映射到四次多项式，存储在X_train_poly4
poly4 = PolynomialFeatures(degree=4)
X_train_poly4 = poly4.fit_transform(X_train)
#以线性回归器为基础，初始化回归模型，尽管特征的维度有所提升，但是模型基础仍然是线性模型
regressor_poly4 = LinearRegression()
#对4次多项式回归模型进行训练
regressor_poly4.fit(X_train_poly4,y_train)
#从新映射绘图用x轴采样数据
xx_poly4 = poly4.transform(xx)
#使用4次多项式回归模型对应x轴采样数据进行回归预测
yy_poly4 = regressor_poly4.predict(xx_poly4)
'''
分别对训练数据点、线性回归直线、4次多项式回归曲线进行作图
'''
plt.scatter(X_train,y_train)
plt1,=plt.plot(xx,yy,label='Degree=1')
plt2,=plt.plot(xx,yy_poly2,label='Degree=2')
plt4,=plt.plot(xx,yy_poly4,label='Degree=4')
plt.axis([0,25,0,25])
plt.xlabel('Diameter of Pizza')
plt.ylabel('Price of Pizza')
plt.legend(handles=[plt1,plt2,plt4])
plt.show()
'''
输出4次多项式回归模型在训练样本上的R-squared值
'''
print('the R-squared value of Polynomial Regressor(Degree=4) performing on training data is:',
regressor_poly4.score(X_train_poly4,y_train))[[ 0. ]
[ 0.26262626]
[ 0.52525253]
[ 0.78787879]
[ 1.05050505]
[ 1.31313131]
[ 1.57575758]
[ 1.83838384]
[ 2.1010101 ]
[ 2.36363636]
[ 2.62626263]
[ 2.88888889]
[ 3.15151515]
[ 3.41414141]
[ 3.67676768]
[ 3.93939394]
[ 4.2020202 ]
[ 4.46464646]
[ 4.72727273]
[ 4.98989899]
[ 5.25252525]
[ 5.51515152]
[ 5.77777778]
[ 6.04040404]
[ 6.3030303 ]
[ 6.56565657]
[ 6.82828283]
[ 7.09090909]
[ 7.35353535]
[ 7.61616162]
[ 7.87878788]
[ 8.14141414]
[ 8.4040404 ]
[ 8.66666667]
[ 8.92929293]
[ 9.19191919]
[ 9.45454545]
[ 9.71717172]
[ 9.97979798]
[ 10.24242424]
[ 10.50505051]
[ 10.76767677]
[ 11.03030303]
[ 11.29292929]
[ 11.55555556]
[ 11.81818182]
[ 12.08080808]
[ 12.34343434]
[ 12.60606061]
[ 12.86868687]
[ 13.13131313]
[ 13.39393939]
[ 13.65656566]
[ 13.91919192]
[ 14.18181818]
[ 14.44444444]
[ 14.70707071]
[ 14.96969697]
[ 15.23232323]
[ 15.49494949]
[ 15.75757576]
[ 16.02020202]
[ 16.28282828]
[ 16.54545455]
[ 16.80808081]
[ 17.07070707]
[ 17.33333333]
[ 17.5959596 ]
[ 17.85858586]
[ 18.12121212]
[ 18.38383838]
[ 18.64646465]
[ 18.90909091]
[ 19.17171717]
[ 19.43434343]
[ 19.6969697 ]
[ 19.95959596]
[ 20.22222222]
[ 20.48484848]
[ 20.74747475]
[ 21.01010101]
[ 21.27272727]
[ 21.53535354]
[ 21.7979798 ]
[ 22.06060606]
[ 22.32323232]
[ 22.58585859]
[ 22.84848485]
[ 23.11111111]
[ 23.37373737]
[ 23.63636364]
[ 23.8989899 ]
[ 24.16161616]
[ 24.42424242]
[ 24.68686869]
[ 24.94949495]
[ 25.21212121]
[ 25.47474747]
[ 25.73737374]
[ 26. ]]
the R-squared value of Linear Regressor performing on the training data is: 0.910001596424
the R-squared value of Polynomial Regressor(Degree=2) performing on training data is: 0.98164216396
the R-squared value of Polynomial Regressor(Degree=4) performing on training data is: 1.0
2.2 L1 Regularization
from sklearn.linear_model import LinearRegression
'''
使用线性回归模型在比萨训练样本上进行拟合
'''
#输入训练样本的特征及其目标值，分别存储在变量X_train与y_train
X_train = [[6],[8],[10],[14],[18]]
y_train = [[7],[9],[13],[17.5],[18]]
X_test = [[6],[8],[11],[16]]
y_test = [[8],[12],[15],[18]]
#使用默认配置初始化线性回归模型
regressor = LinearRegression()
#直接以比萨的半径作为特征训练模型
regressor.fit(X_train,y_train)
import numpy as np
#在x轴上从0到25均匀采样100个数据点
xx = np.linspace(0,26,100)
# print(xx.shape[0])
xx = xx.reshape(xx.shape[0],1)
print(xx)
#以上述100个数据点作为基准，预测回归直线
yy = regressor.predict(xx)
#对回归预测到直线进行作图
import matplotlib.pyplot as plt
plt.scatter(X_train,y_train) #画出样本散点图
plt1,=plt.plot(xx,yy,label="Degree=1") #画出回归预测直线
plt.axis([0,25,0,25]) #设置坐标轴的范围
plt.xlabel('Diameter of Pizza')
plt.ylabel('Price of pizza')
plt.legend(handles=[plt1])
plt.show()
'''
Lasso模型在4次多项式特征上拟合的表现,等同于在线性回归目标加入L1范数正则化，避免参数过拟合
'''
'''
使用4次多项式回归模型在披萨训练样本上进行拟合
'''
from sklearn.preprocessing import PolynomialFeatures
#使用PolynomialFratures将x特征映射到四次多项式，存储在X_train_poly4
poly4 = PolynomialFeatures(degree=4)
X_train_poly4 = poly4.fit_transform(X_train)
#以线性回归器为基础，初始化回归模型，尽管特征的维度有所提升，但是模型基础仍然是线性模型
regressor_poly4 = LinearRegression()
#对4次多项式回归模型进行训练
regressor_poly4.fit(X_train_poly4,y_train)
X_test_poly4 = poly4.transform(X_test)
print('the accuracy is:',regressor_poly4.score(X_test_poly4,y_test))
#从sklearn.linear_model中导入Lasso
from sklearn.linear_model import Lasso
#从使用配置初始化Lasso
lasso_poly4 = Lasso()
#从使用Lasso对4次多项式特征进行拟合
lasso_poly4.fit(X_train_poly4,y_train)
X_test_poly4_lasso = poly4.transform(X_test)
print('the accuracy is:',lasso_poly4.score(X_test_poly4_lasso,y_test))[[ 0. ]
[ 0.26262626]
[ 0.52525253]
[ 0.78787879]
[ 1.05050505]
[ 1.31313131]
[ 1.57575758]
[ 1.83838384]
[ 2.1010101 ]
[ 2.36363636]
[ 2.62626263]
[ 2.88888889]
[ 3.15151515]
[ 3.41414141]
[ 3.67676768]
[ 3.93939394]
[ 4.2020202 ]
[ 4.46464646]
[ 4.72727273]
[ 4.98989899]
[ 5.25252525]
[ 5.51515152]
[ 5.77777778]
[ 6.04040404]
[ 6.3030303 ]
[ 6.56565657]
[ 6.82828283]
[ 7.09090909]
[ 7.35353535]
[ 7.61616162]
[ 7.87878788]
[ 8.14141414]
[ 8.4040404 ]
[ 8.66666667]
[ 8.92929293]
[ 9.19191919]
[ 9.45454545]
[ 9.71717172]
[ 9.97979798]
[ 10.24242424]
[ 10.50505051]
[ 10.76767677]
[ 11.03030303]
[ 11.29292929]
[ 11.55555556]
[ 11.81818182]
[ 12.08080808]
[ 12.34343434]
[ 12.60606061]
[ 12.86868687]
[ 13.13131313]
[ 13.39393939]
[ 13.65656566]
[ 13.91919192]
[ 14.18181818]
[ 14.44444444]
[ 14.70707071]
[ 14.96969697]
[ 15.23232323]
[ 15.49494949]
[ 15.75757576]
[ 16.02020202]
[ 16.28282828]
[ 16.54545455]
[ 16.80808081]
[ 17.07070707]
[ 17.33333333]
[ 17.5959596 ]
[ 17.85858586]
[ 18.12121212]
[ 18.38383838]
[ 18.64646465]
[ 18.90909091]
[ 19.17171717]
[ 19.43434343]
[ 19.6969697 ]
[ 19.95959596]
[ 20.22222222]
[ 20.48484848]
[ 20.74747475]
[ 21.01010101]
[ 21.27272727]
[ 21.53535354]
[ 21.7979798 ]
[ 22.06060606]
[ 22.32323232]
[ 22.58585859]
[ 22.84848485]
[ 23.11111111]
[ 23.37373737]
[ 23.63636364]
[ 23.8989899 ]
[ 24.16161616]
[ 24.42424242]
[ 24.68686869]
[ 24.94949495]
[ 25.21212121]
[ 25.47474747]
[ 25.73737374]
[ 26. ]]
the accuracy is: 0.809588079577
the accuracy is: 0.83889268736
C:\Users\xxz\Anaconda3\lib\site-packages\sklearn\linear_model\coordinate_descent.py:466: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations
ConvergenceWarning)
2.3 L2 Regularization
from sklearn.linear_model import LinearRegression
'''
使用线性回归模型在比萨训练样本上进行拟合
'''
#输入训练样本的特征及其目标值，分别存储在变量X_train与y_train
X_train = [[6],[8],[10],[14],[18]]
y_train = [[7],[9],[13],[17.5],[18]]
X_test = [[6],[8],[11],[16]]
y_test = [[8],[12],[15],[18]]
#使用默认配置初始化线性回归模型
regressor = LinearRegression()
#直接以比萨的半径作为特征训练模型
regressor.fit(X_train,y_train)
import numpy as np
#在x轴上从0到25均匀采样100个数据点
xx = np.linspace(0,26,100)
# print(xx.shape[0])
xx = xx.reshape(xx.shape[0],1)
print(xx)
#以上述100个数据点作为基准，预测回归直线
yy = regressor.predict(xx)
#对回归预测到直线进行作图
import matplotlib.pyplot as plt
plt.scatter(X_train,y_train) #画出样本散点图
plt1,=plt.plot(xx,yy,label="Degree=1") #画出回归预测直线
plt.axis([0,25,0,25]) #设置坐标轴的范围
plt.xlabel('Diameter of Pizza')
plt.ylabel('Price of pizza')
plt.legend(handles=[plt1])
plt.show()
'''
Ridge模型在4次多项式特征上拟合的表现,等同于在线性回归目标加入L2范数正则化，避免参数过拟合
'''
'''
使用4次多项式回归模型在披萨训练样本上进行拟合
'''
from sklearn.preprocessing import PolynomialFeatures
#使用PolynomialFratures将x特征映射到四次多项式，存储在X_train_poly4
poly4 = PolynomialFeatures(degree=4)
X_train_poly4 = poly4.fit_transform(X_train)
#以线性回归器为基础，初始化回归模型，尽管特征的维度有所提升，但是模型基础仍然是线性模型
regressor_poly4 = LinearRegression()
#对4次多项式回归模型进行训练
regressor_poly4.fit(X_train_poly4,y_train)
X_test_poly4 = poly4.transform(X_test)
print('the accuracy is:',regressor_poly4.score(X_test_poly4,y_test))
#输出上述参数的平方和，验证参数之前的巨大差异
print(regressor_poly4.coef_)
print(np.sum(regressor_poly4.coef_ ** 2))
#从sklearn.linear_model中导入Lasso
from sklearn.linear_model import Ridge
#从使用配置初始化Lasso
Ridge_poly4 = Ridge()
#从使用Lasso对4次多项式特征进行拟合
Ridge_poly4.fit(X_train_poly4,y_train)
X_test_poly4_Ridge = poly4.transform(X_test)
print('the accuracy is:',Ridge_poly4.score(X_test_poly4_Ridge,y_test))
#输出上述参数的平方和，验证参数之前的巨大差异
print(Ridge_poly4.coef_)
print(np.sum(Ridge_poly4.coef_ ** 2))[[ 0. ]
[ 0.26262626]
[ 0.52525253]
[ 0.78787879]
[ 1.05050505]
[ 1.31313131]
[ 1.57575758]
[ 1.83838384]
[ 2.1010101 ]
[ 2.36363636]
[ 2.62626263]
[ 2.88888889]
[ 3.15151515]
[ 3.41414141]
[ 3.67676768]
[ 3.93939394]
[ 4.2020202 ]
[ 4.46464646]
[ 4.72727273]
[ 4.98989899]
[ 5.25252525]
[ 5.51515152]
[ 5.77777778]
[ 6.04040404]
[ 6.3030303 ]
[ 6.56565657]
[ 6.82828283]
[ 7.09090909]
[ 7.35353535]
[ 7.61616162]
[ 7.87878788]
[ 8.14141414]
[ 8.4040404 ]
[ 8.66666667]
[ 8.92929293]
[ 9.19191919]
[ 9.45454545]
[ 9.71717172]
[ 9.97979798]
[ 10.24242424]
[ 10.50505051]
[ 10.76767677]
[ 11.03030303]
[ 11.29292929]
[ 11.55555556]
[ 11.81818182]
[ 12.08080808]
[ 12.34343434]
[ 12.60606061]
[ 12.86868687]
[ 13.13131313]
[ 13.39393939]
[ 13.65656566]
[ 13.91919192]
[ 14.18181818]
[ 14.44444444]
[ 14.70707071]
[ 14.96969697]
[ 15.23232323]
[ 15.49494949]
[ 15.75757576]
[ 16.02020202]
[ 16.28282828]
[ 16.54545455]
[ 16.80808081]
[ 17.07070707]
[ 17.33333333]
[ 17.5959596 ]
[ 17.85858586]
[ 18.12121212]
[ 18.38383838]
[ 18.64646465]
[ 18.90909091]
[ 19.17171717]
[ 19.43434343]
[ 19.6969697 ]
[ 19.95959596]
[ 20.22222222]
[ 20.48484848]
[ 20.74747475]
[ 21.01010101]
[ 21.27272727]
[ 21.53535354]
[ 21.7979798 ]
[ 22.06060606]
[ 22.32323232]
[ 22.58585859]
[ 22.84848485]
[ 23.11111111]
[ 23.37373737]
[ 23.63636364]
[ 23.8989899 ]
[ 24.16161616]
[ 24.42424242]
[ 24.68686869]
[ 24.94949495]
[ 25.21212121]
[ 25.47474747]
[ 25.73737374]
[ 26. ]]
the accuracy is: 0.809588079577
[[ 0.00000000e+00 -2.51739583e+01 3.68906250e+00 -2.12760417e-01
4.29687500e-03]]
647.382645737
the accuracy is: 0.837420175937
[[ 0. -0.00492536 0.12439632 -0.00046471 -0.00021205]]
0.0154989652036
3. 模型检验
'''
模型检验主要有两种方法:
(1) 留一验证
(2) 交叉验证
''''\n模型检验主要有两种方法:\n(1) 留一验证\n(2) 交叉验证\n'
4. 超参数搜索
4.1 Grid Searching
import numpy as np
'''
网格搜索(Grid Searching)对多种超参数组合的空间进行暴力搜索，每一套超参数组合被代入到学习函数中作为新的模型，并且比较新模型之间的性能，
每个模型都会采用交叉验证的方法在多组相同的训练和开发数据集下进行评估。
'''
'''
使用单线程对文本分类的朴素贝叶斯模型的超参数组合执行网络搜索
'''
#从sklearn.datasets中导入20类新闻文本抓取器
from sklearn.datasets import fetch_20newsgroups
#使用新闻抓取器从互联网上下载所有数据，并且存储在变量news中
news = fetch_20newsgroups(subset='all')
#从sklearn.cross_validation导入train_test_split用来分割数据
from sklearn.cross_validation import train_test_split
#对前3000条新闻文本进行数据分割，25%文本用于未来测试
X_train,X_test,y_train,y_test = train_test_split(news.data[:3000],news.target[:3000],test_size=0.25,random_state=33)
#导入支持向量机（分类）模型
from sklearn.svm import SVC
#导入TfidfVectorizer文本抽取器
from sklearn.feature_extraction.text import TfidfVectorizer
#导入Pipeline
from sklearn.pipeline import Pipeline
#使用Pipeline简化系统搭建流程，将文本抽取与分类器模型串联起来
clf = Pipeline([('vect',TfidfVectorizer(stop_words='english',analyzer='word')),('svc',SVC())])
#这里需要试验的2个超参数的个数分别是4、3，svc__gamma的参数共有10^-2,10^-1……，这样我们一共有12种的超参数组合，12个不同参数下的模型
parameters = {'svc__gamma':np.logspace(-2,1,4),'svc__C':np.logspace(-1,1,3)}
#从sklearn.grid_search中导入网络搜索模块GridSearchCV
from sklearn.grid_search import GridSearchCV
#将12组参数组合以及初始化的Pipline包括3折交叉验证的要求全部告知GridSearchCV
gs = GridSearchCV(clf,parameters,verbose=2,refit=True,cv=3)
#执行单线程网络搜索
time_ = gs.fit(X_train,y_train)
print(gs.best_params_, gs.best_score_)
#输出最佳模型在测试集上的准确性
print(gs.score(X_test,y_test))Fitting 3 folds for each of 12 candidates, totalling 36 fits
[CV] svc__C=0.1, svc__gamma=0.01 .....................................
[CV] ............................ svc__C=0.1, svc__gamma=0.01 - 7.9s
[CV] svc__C=0.1, svc__gamma=0.01 .....................................
[CV] ............................ svc__C=0.1, svc__gamma=0.01 - 6.0s
[CV] svc__C=0.1, svc__gamma=0.01 .....................................
[CV] ............................ svc__C=0.1, svc__gamma=0.01 - 6.3s
[CV] svc__C=0.1, svc__gamma=0.1 ......................................
[CV] ............................. svc__C=0.1, svc__gamma=0.1 - 5.9s
[CV] svc__C=0.1, svc__gamma=0.1 ......................................
[CV] ............................. svc__C=0.1, svc__gamma=0.1 - 5.9s
[CV] svc__C=0.1, svc__gamma=0.1 ......................................
[CV] ............................. svc__C=0.1, svc__gamma=0.1 - 6.2s
[CV] svc__C=0.1, svc__gamma=1.0 ......................................
[CV] ............................. svc__C=0.1, svc__gamma=1.0 - 6.1s
[CV] svc__C=0.1, svc__gamma=1.0 ......................................
[CV] ............................. svc__C=0.1, svc__gamma=1.0 - 6.0s
[CV] svc__C=0.1, svc__gamma=1.0 ......................................
[CV] ............................. svc__C=0.1, svc__gamma=1.0 - 6.1s
[CV] svc__C=0.1, svc__gamma=10.0 .....................................
[CV] ............................ svc__C=0.1, svc__gamma=10.0 - 6.3s
[CV] svc__C=0.1, svc__gamma=10.0 .....................................
[CV] ............................ svc__C=0.1, svc__gamma=10.0 - 6.3s
[CV] svc__C=0.1, svc__gamma=10.0 .....................................
[CV] ............................ svc__C=0.1, svc__gamma=10.0 - 6.7s
[CV] svc__C=1.0, svc__gamma=0.01 .....................................
[CV] ............................ svc__C=1.0, svc__gamma=0.01 - 6.3s
[CV] svc__C=1.0, svc__gamma=0.01 .....................................
[CV] ............................ svc__C=1.0, svc__gamma=0.01 - 6.1s
[CV] svc__C=1.0, svc__gamma=0.01 .....................................
[CV] ............................ svc__C=1.0, svc__gamma=0.01 - 6.2s
[CV] svc__C=1.0, svc__gamma=0.1 ......................................
[CV] ............................. svc__C=1.0, svc__gamma=0.1 - 5.8s
[CV] svc__C=1.0, svc__gamma=0.1 ......................................
[CV] ............................. svc__C=1.0, svc__gamma=0.1 - 5.9s
[CV] svc__C=1.0, svc__gamma=0.1 ......................................
[CV] ............................. svc__C=1.0, svc__gamma=0.1 - 6.2s
[CV] svc__C=1.0, svc__gamma=1.0 ......................................
[CV] ............................. svc__C=1.0, svc__gamma=1.0 - 6.0s
[CV] svc__C=1.0, svc__gamma=1.0 ......................................
[CV] ............................. svc__C=1.0, svc__gamma=1.0 - 6.1s
[CV] svc__C=1.0, svc__gamma=1.0 ......................................
[CV] ............................. svc__C=1.0, svc__gamma=1.0 - 6.3s
[CV] svc__C=1.0, svc__gamma=10.0 .....................................
[CV] ............................ svc__C=1.0, svc__gamma=10.0 - 6.4s
[CV] svc__C=1.0, svc__gamma=10.0 .....................................
[CV] ............................ svc__C=1.0, svc__gamma=10.0 - 6.4s
[CV] svc__C=1.0, svc__gamma=10.0 .....................................
[CV] ............................ svc__C=1.0, svc__gamma=10.0 - 6.4s
[CV] svc__C=10.0, svc__gamma=0.01 ....................................
[CV] ........................... svc__C=10.0, svc__gamma=0.01 - 6.3s
[CV] svc__C=10.0, svc__gamma=0.01 ....................................
[CV] ........................... svc__C=10.0, svc__gamma=0.01 - 6.1s
[CV] svc__C=10.0, svc__gamma=0.01 ....................................
[CV] ........................... svc__C=10.0, svc__gamma=0.01 - 6.1s
[CV] svc__C=10.0, svc__gamma=0.1 .....................................
[CV] ............................ svc__C=10.0, svc__gamma=0.1 - 6.0s
[CV] svc__C=10.0, svc__gamma=0.1 .....................................
[CV] ............................ svc__C=10.0, svc__gamma=0.1 - 6.0s
[CV] svc__C=10.0, svc__gamma=0.1 .....................................
[CV] ............................ svc__C=10.0, svc__gamma=0.1 - 6.2s
[CV] svc__C=10.0, svc__gamma=1.0 .....................................
[CV] ............................ svc__C=10.0, svc__gamma=1.0 - 6.0s
[CV] svc__C=10.0, svc__gamma=1.0 .....................................
[CV] ............................ svc__C=10.0, svc__gamma=1.0 - 6.1s
[CV] svc__C=10.0, svc__gamma=1.0 .....................................
[CV] ............................ svc__C=10.0, svc__gamma=1.0 - 6.2s
[CV] svc__C=10.0, svc__gamma=10.0 ....................................
[CV] ........................... svc__C=10.0, svc__gamma=10.0 - 6.2s
[CV] svc__C=10.0, svc__gamma=10.0 ....................................
[CV] ........................... svc__C=10.0, svc__gamma=10.0 - 6.3s
[CV] svc__C=10.0, svc__gamma=10.0 ....................................
[CV] ........................... svc__C=10.0, svc__gamma=10.0 - 6.3s
[Parallel(n_jobs=1)]: Done 36 out of 36 | elapsed: 3.8min finished
{'svc__C': 10.0, 'svc__gamma': 0.10000000000000001} 0.790666666667
0.822666666667
4.2 Parallel Grid Search
import numpy as np
'''
网格搜索(Grid Searching)对多种超参数组合的空间进行暴力搜索，每一套超参数组合被代入到学习函数中作为新的模型，并且比较新模型之间的性能，
每个模型都会采用交叉验证的方法在多组相同的训练和开发数据集下进行评估。
'''
'''
使用多线程对文本分类的朴素贝叶斯模型的超参数组合执行网络搜索
'''
#从sklearn.datasets中导入20类新闻文本抓取器
from sklearn.datasets import fetch_20newsgroups
import numpy as np
#使用新闻抓取器从互联网上下载所有数据，并且存储在变量news中
news = fetch_20newsgroups(subset='all')
#从sklearn.cross_validation导入train_test_split用来分割数据
from sklearn.cross_validation import train_test_split
#对前3000条新闻文本进行数据分割，25%文本用于未来测试
X_train,X_test,y_train,y_test = train_test_split(news.data[:3000],news.target[:3000],test_size=0.25,random_state=33)
#导入支持向量机（分类）模型
from sklearn.svm import SVC
#导入TfidfVectorizer文本抽取器
from sklearn.feature_extraction.text import TfidfVectorizer
#导入Pipeline
from sklearn.pipeline import Pipeline
#使用Pipeline简化系统搭建流程，将文本抽取与分类器模型串联起来
clf = Pipeline([('vect',TfidfVectorizer(stop_words='english',analyzer='word')),('svc',SVC())])
#这里需要试验的2个超参数的个数分别是4、3，svc__gamma的参数共有10^-2,10^-1……，这样我们一共有12种的超参数组合，12个不同参数下的模型
parameters = {'svc__gamma':np.logspace(-2,1,4),'svc__C':np.logspace(-1,1,3)}
#从sklearn.grid_search中导入网络搜索模块GridSearchCV
from sklearn.grid_search import GridSearchCV
if __name__ == '__main__':
#将12组参数组合以及初始化的Pipline包括3折交叉验证的要求全部告知GridSearchCV
#初始化配置并行网络搜索，n_jobs=-1代表使用该计算机全部的CPU
gs = GridSearchCV(clf,parameters,verbose=2,refit=True,cv=3,n_jobs=-1)
#执行多线程并行网络搜索
time_ = gs.fit(X_train,y_train)
print(gs.best_params_, gs.best_score_)
#输出最佳模型在测试集上的准确性
print(gs.score(X_test,y_test))Fitting 3 folds for each of 12 candidates, totalling 36 fits
[Parallel(n_jobs=-1)]: Done 36 out of 36 | elapsed: 59.9s finished
{'svc__C': 10.0, 'svc__gamma': 0.10000000000000001} 0.790666666667
0.822666666667
扫一扫
8064
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
