来自：http://cloga.info/python/2014/02/07/classify_use_Sklearn/#wat_e_12612920-6fe4-464e-a2b0-3b1f13c1a4f6_zss_
加载数据集
这里我使用pandas来加载数据集，数据集采用kaggle的titanic的数据集，下载train.csv。
import pandas as pd
df = pd.read_csv('train.csv')
df = df.fillna(0) #将缺失值都替换为0
df.head()
| PassengerId | Survived | Pclass | Name | Sex | Age | SibSp | Parch | Ticket | Fare | Cabin | Embarked | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 1 | 0 | 3 | Braund, Mr. Owen Harris | male | 22 | 1 | 0 | A/5 21171 | 7.2500 | 0 | S |
| 1 | 2 | 1 | 1 | Cumings, Mrs. John Bradley (Florence Briggs Th... | female | 38 | 1 | 0 | PC 17599 | 71.2833 | C85 | C |
| 2 | 3 | 1 | 3 | Heikkinen, Miss. Laina | female | 26 | 0 | 0 | STON/O2. 3101282 | 7.9250 | 0 | S |
| 3 | 4 | 1 | 1 | Futrelle, Mrs. Jacques Heath (Lily May Peel) | female | 35 | 1 | 0 | 113803 | 53.1000 | C123 | S |
| 4 | 5 | 0 | 3 | Allen, Mr. William Henry | male | 35 | 0 | 0 | 373450 | 8.0500 | 0 | S |
5 rows × 12 columns
len(df)
891
可以看到训练集中共有891条记录，有12个列（其中一列Survived是目标分类）。将数据集分为特征集和目标分类集，两个DataFrame。
exc_cols = [u'PassengerId', u'Survived', u'Name']
cols = [c for c in df.columns if c not in exc_cols]
x = df.ix[:,cols]
y = df['Survived'].values
由于Sklearn为了效率，接受的特征数据类型是dtype=np.float32以便获得最佳的算法效率。因此，对于类别类型的特征就需要转化为向量。Sklearn 提供了DictVectorizer类将类别的特征转化为向量。DictVectorizer接受记录的形式为字典的列表。因此需要用pandas的to_dict方法转 换DataFrame。
from sklearn.feature_extraction import DictVectorizer
v = DictVectorizer()
x = v.fit_transform(x.to_dict(outtype='records')).toarray()
让我们比较一下同一个实例的原始信息及向量化后的结果。
print 'Vectorized:', x[10]
print 'Unvectorized:', v.inverse_transform(x[10])
Vectorized: [ 4. 0. 0. ..., 0. 0. 0.]
Unvectorized: [{'Fare': 16.699999999999999, 'Name=Sandstrom, Miss. Marguerite Rut': 1.0, 'Embarked=S': 1.0, 'Age': 4.0, 'Sex=female': 1.0, 'Parch': 1.0, 'Pclass': 3.0, 'Ticket=PP 9549': 1.0, 'Cabin=G6': 1.0, 'SibSp': 1.0, 'PassengerId': 11.0}]
如果分类的标签也是字符的，那么就还需要用LabelEncoder方法进行转化。
将数据集分成训练集和测试集。
from sklearn.cross_validation import train_test_split
data_train, data_test, target_train, target_test = train_test_split(x, y)
len(data_train)
668
len(data_test)
223
默认是以数据集的25%作为测试集。到这里为止，用于训练和测试的数据集都已经准备好了。
用Sklearn做判别分析
Sklearn训练模型的基本流程
Model = EstimatorObject()
Model.fit(dataset.data, dataset.target)
dataset.data = dataset
dataset.target = labels
Model.predict(dataset.data)
这里选择朴素贝叶斯、决策树、随机森林和SVM来做一个对比。
from sklearn import cross_validation
from sklearn.naive_bayes import GaussianNB
from sklearn import tree
from sklearn.ensemble import RandomForestClassifier
from sklearn import svm
import datetime
estimators = {}
estimators['bayes'] = GaussianNB()
estimators['tree'] = tree.DecisionTreeClassifier()
estimators['forest_100'] = RandomForestClassifier(n_estimators = 100)
estimators['forest_10'] = RandomForestClassifier(n_estimators = 10)
estimators['svm_c_rbf'] = svm.SVC()
estimators['svm_c_linear'] = svm.SVC(kernel='linear')
estimators['svm_linear'] = svm.LinearSVC()
estimators['svm_nusvc'] = svm.NuSVC()
首先是定义各个model所用的算法。
for k in estimators.keys():
start_time = datetime.datetime.now()
print '----%s----' % k
estimators[k] = estimators[k].fit(data_train, target_train)
pred = estimators[k].predict(data_test)
print("%s Score: %0.2f" % (k, estimators[k].score(data_test, target_test)))
scores = cross_validation.cross_val_score(estimators[k], data_test, target_test, cv=5)
print("%s Cross Avg. Score: %0.2f (+/- %0.2f)" % (k, scores.mean(), scores.std() * 2))
end_time = datetime.datetime.now()
time_spend = end_time - start_time
print("%s Time: %0.2f" % (k, time_spend.total_seconds()))
----svm_c_rbf----
svm_c_rbf Score: 0.63
svm_c_rbf Cross Avg. Score: 0.54 (+/- 0.18)
svm_c_rbf Time: 1.67
----tree----
tree Score: 0.81
tree Cross Avg. Score: 0.75 (+/- 0.09)
tree Time: 0.90
----forest_10----
forest_10 Score: 0.83
forest_10 Cross Avg. Score: 0.80 (+/- 0.10)
forest_10 Time: 0.56
----forest_100----
forest_100 Score: 0.84
forest_100 Cross Avg. Score: 0.80 (+/- 0.14)
forest_100 Time: 5.38
----svm_linear----
svm_linear Score: 0.74
svm_linear Cross Avg. Score: 0.65 (+/- 0.18)
svm_linear Time: 0.15
----svm_nusvc----
svm_nusvc Score: 0.63
svm_nusvc Cross Avg. Score: 0.55 (+/- 0.21)
svm_nusvc Time: 1.62
----bayes----
bayes Score: 0.44
bayes Cross Avg. Score: 0.47 (+/- 0.07)
bayes Time: 0.16
----svm_c_linear----
svm_c_linear Score: 0.83
svm_c_linear Cross Avg. Score: 0.79 (+/- 0.14)
svm_c_linear Time: 465.57
这里通过算法的score方法及cross_validation来计算预测的准确性。
可以看到准确性比较高的算法需要的时间也会增加。性价比较高的算法是随机森林。 让我们用kaggle给出的test.csv的数据集测试一下。
test = pd.read_csv('test.csv')
test = test.fillna(0)
test_d = test.to_dict(outtype='records')
test_vec = v.transform(test_d).toarray()
这里需要注意的是test的数据也需要经过同样的DictVectorizer转换。
for k in estimators.keys():
estimators[k] = estimators[k].fit(x, y)
pred = estimators[k].predict(test_vec)
test['Survived'] = pred
test.to_csv(k + '.csv', cols=['Survived', 'PassengerId'], index=False)
好了，向Kaggle提交你的结果吧~
你可以查看本文的ipython notebook版本