1 from sklearn.linear_model import LogisticRegression #逻辑回归
2 from sklearn import cross_validation
3 alg = LogisticRegression(random_state=1)
4 scores = cross_validation.cross_val_score(alg, titanic[predictors],titanic['Survived'],cv=3)
5 print scores.mean()

 1 from sklearn.ensemble import RandomForestClassifier
 2 from sklearn import cross_validation
 3 predictions = ['Pclass','Sex','Age','SibSp','Parch','Fare','Embarked']
 4 # Initialize our algorithm with the default paramters
 5 # random_state = 1 表示此处代码多运行几次得到的随机值都是一样的，如果不设置，两次执行的随机值是不一样的
 6 # n_estimators  指定有多少颗决策树，树的分裂的条件是:
 7 # min_samples_split 代表样本不停的分裂，某一个节点上的样本如果只有2个了 ，就不再继续分裂了
 8 # min_samples_leaf 是控制叶子节点的最小个数
 9 alg = RandomForestClassifier(random_state=1,n_estimators=100,min_samples_split=4,min_samples_leaf=2)
10 #进行交叉验证
11 kf = cross_validation.KFold(titanic.shape[0],n_folds=3,random_state=1)
12 scores = cross_validation.cross_val_score(alg,titanic[predictors],titanic['Survived'],cv=kf)
13 print scores.mean()

 1 # ## 关于特征提取问题 (非常关键)
 2 # - 尽可能多的提取特征
 3 # - 看不同特征的效果
 4 # - 特征提取是数据挖掘里很- 要的一部分
 5 # - 以上使用的特征都是数据里已经有的了，在真实的数据挖掘里我们常常没有合适的特征，需要我们自己取提取
 6 ① 把多个特征组合成一个特征
 7 titanic['Familysize'] = titanic['SibSp'] + titanic['Parch'] #家庭总共多少人
 8 titanic['NameLength'] = titanic['Name'].apply(lambda x: len(x)) #名字的长度
 9 import re
10
11 def get_title(name):
12     title_reserch = re.search('([A-Za-z]+)\.',name)
13     if title_reserch:
14         return title_reserch.group(1)
15     return ""
16 titles = titanic['Name'].apply(get_title)
17 #print pandas.value_counts(titles)
18
19 #将称号转换成数值表示
20 title_mapping = {"Mr":1,"Miss":2,"Mrs":3,"Master":4,"Dr":5,"Rev":6,"Col":7,"Major":8,"Mlle":9,"Countess":10,"Ms":11,"Lady":12,"Jonkheer":13,"Don":14,"Mme":15,"Capt":16,"Sir":17}
21 for k,v in title_mapping.items():
22     titles[titles==k] = v
23     #print (pandas.value_counts(titles))
24 titanic["titles"] = titles #添加title特征

 1 # 在竞赛中常用的耍赖的办法:集成多种算法，取最后每种算法的平均值，来减少过拟合
 2 from sklearn.ensemble import GradientBoostingClassifier
 3 import numpy as np
 4 # GradientBoostingClassifier也是一种随机森林的算法，可以集成多个弱分类器，然后变成强分类器
 5 algorithas = [
 6         [GradientBoostingClassifier(random_state=1,n_estimators=25,max_depth=3),['Pclass','Sex','Age','SibSp','Parch','Fare','Embarked','Familysize','NameLength','titles']],
 7         [LogisticRegression(random_state=1),['Pclass','Sex','Age','SibSp','Parch','Fare','Embarked','Familysize','NameLength','titles']]
 8         ]
 9 kf = KFold(titanic.shape[0],n_folds=3,random_state=1)
10 predictions = []
11 for train, test in kf:
12    train_target = titanic['Survived'].iloc[train]
13    full_test_predictions = []
14    for alg,predictors in algorithas:
15        alg.fit(titanic[predictors].iloc[train,:],train_target)
16        test_prediction = alg.predict_proba(titanic[predictors].iloc[test,:].astype(float))[:,1]
17        full_test_predictions.append(test_prediction)
18    test_predictions = (full_test_predictions[0] + full_test_predictions[1]) / 2
19    test_predictions[test_predictions >.5] = 1
20    test_predictions[test_predictions <=.5] = 0
21    predictions.append(test_predictions)
22 predictions = np.concatenate(predictions,axis=0)
23 accury = sum(predictions[predictions == titanic['Survived']]) / len(predictions)#测试准确率
24 print accury