from sklearn.model_selection import train_test_split, GridSearchCV from sklearn.neighbors import KNeighborsClassifier from sklearn.preprocessing import StandardScaler from sklearn.feature_extraction.text import TfidfVectorizer from sklearn.naive_bayes import MultinomialNB from sklearn.metrics import classification_report from sklearn.feature_extraction import DictVectorizer from sklearn.tree import DecisionTreeClassifier, export_graphviz from sklearn.ensemble import RandomForestClassifier import pandas as pd import numpy as np import matplotlib.pyplot as plt import re from sklearn import preprocessing from sklearn.feature_extraction import DictVectorizer from sklearn.linear_model import LogisticRegression plt.rcParams['font.sans-serif']=['SimHei'] #显示中文标签 plt.rcParams['axes.unicode_minus']=False

导入数据：

data_train = pd.read_csv("F:/kaggle数据/titanic/train.csv") data_test = pd.read_csv("F:/kaggle数据/titanic/test.csv")

查看缺失信息：

""" 训练集中"Age"、"Cabin"、"Embarked"缺失 测试集中"Age"、"Fare"、"Cabin"缺失 测试集中无"Survived"标签值 """

缺失值处理方法：

""" 1.直接删除此特征（缺损数据太多的情况，防止引入噪声） 2.直接删除缺损数据的样本（~土豪操作~只用于训练数据集，且样本量较大，缺损数据样本较少的情况） 3.直接将有无数值作为新的特征（数据缺失较多，且数据有无本身是对预测是一个有用的特征） 4.中值或均值回补（缺失数据较多，不想损失此较多训练数据，特征又比较重要的情况，是比较常用的方法） 5.参考其他特征，利用与此特征的相关性编写算法回补数据（~大神级操作~回补的准确性可能会比较高一些，但实现过程复杂） """ data_train.isnull().any() 结果： PassengerId False Survived False Pclass False Name False Sex False Age True SibSp False Parch False Ticket False Fare False Cabin True Embarked True dtype: bool data_test.isnull().any() 结果： PassengerId False Pclass False Name False Sex False Age True SibSp False Parch False Ticket False Fare True Cabin True Embarked False dtype: bool data_train.info(null_counts=True) 结果： <class 'pandas.core.frame.DataFrame'> RangeIndex: 891 entries, 0 to 890 Data columns (total 12 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 PassengerId 891 non-null int64 1 Survived 891 non-null int64 2 Pclass 891 non-null int64 3 Name 891 non-null object 4 Sex 891 non-null object 5 Age 714 non-null float64 6 SibSp 891 non-null int64 7 Parch 891 non-null int64 8 Ticket 891 non-null object 9 Fare 891 non-null float64 10 Cabin 204 non-null object 11 Embarked 889 non-null object dtypes: float64(2), int64 data_test.info(null_counts=True) 结果： <class 'pandas.core.frame.DataFrame'> RangeIndex: 418 entries, 0 to 417 Data columns (total 11 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 PassengerId 418 non-null int64 1 Pclass 418 non-null int64 2 Name 418 non-null object 3 Sex 418 non-null object 4 Age 332 non-null float64 5 SibSp 418 non-null int64 6 Parch 418 non-null int64 7 Ticket 418 non-null object 8 Fare 417 non-null float64 9 Cabin 91 non-null object 10 Embarked 418 non-null object dtypes: float64(2), int64(4), object(5) memory usage: 36.0+ KB

一、查看数据信息

1、船舱等级与生还者的关系

1） 使用不同分类下的value_counts()

bottom = np.zeros(3) x = np.sort(data_train["Pclass"].unique()) for i in range(2): y = data_train[data_train["Survived"]==i].groupby(["Pclass"])["Survived"].value_counts() plt.bar(x,y,bottom=bottom,width=0.5,label="{}".format(i)) for xi,yi,zi in zip(x,bottom,y): plt.text(xi,((yi+zi)+yi)/2,zi,ha="center",va="center") bottom += y.values plt.legend() plt.title("船舱等级与生还的关系") plt.show()

2）直接挑选不同Pclass类别下的Survived分类数,并统计频数合并

h1 = data_train[data_train["Pclass"]==1]["Survived"].value_counts().reindex(index=(0,1)) h2 = data_train[data_train["Pclass"]==2]["Survived"].value_counts() h3 = data_train[data_train["Pclass"]==3]["Survived"].value_counts() pd.DataFrame([list(h1),list(h2),list(h3)],index=(1,2,3),columns=(0,1)) pd.DataFrame([np.array(h1),np.array(h2),np.array(h3)],index=(1,2,3),columns=(1,2)) merge = pd.DataFrame({0:data_train[data_train['Survived'] == 0]["Pclass"].value_counts(),1:data_train[data_train['Survived'] == 1]["Pclass"].value_counts()}) merge

3）使用 pd.crosstab直接进行频数统计

Pclass_Survived = pd.crosstab(index=data_train["Pclass"], columns=data_train["Survived"]) Pclass_Survived.plot(kind="bar",stacked=True) X = np.arange(3) bottom= np.zeros(3) for i in range(2): Z= Pclass_Survived[i] for x,y,z in zip(X,bottom,Z): plt.text(x,((y+z)+y)/2,z,ha="center",va="center") bottom += Z plt.title("船舱等级与生还的关系") plt.show()

2、头衔与生还的关系

data_train['Appellation'] = data_train["Name"].apply(lambda x : re.search("\w+\.",x).group()).str.replace(".","") data_train['Appellation'].replace(['Capt','Col','Countess','Don','Dr','Jonkheer','Lady','Major','Rev','Sir'], 'Rare',inplace=True) data_train['Appellation'].replace(['Mlle','Ms'], 'Miss',inplace=True) data_train['Appellation'].replace('Mme', 'Mrs',inplace=True) data_train.Appellation.unique() Appellation_Survived = pd.crosstab(index = data_train["Appellation"], columns = data_train["Survived"]) Appellation_Survived.plot(kind="bar") plt.xticks(ticks=np.arange(5),labels=Appellation_Survived.index,rotation=360) plt.title("头衔与生还的关系") plt.show()

3、性别和生还的关系

Sex_Survived = pd.crosstab(data_train["Sex"],data_train["Survived"]) Sex_Survived.plot(kind="bar") plt.xticks(ticks=np.arange(2),labels=(Sex_Survived.index),rotation=360) plt.title("性别与生还者的关系") plt.show()

4、兄弟姐妹配偶数量和生还的关系

Sibsp_Survived = pd.crosstab(data_train["SibSp"],data_train["Survived"]) Sibsp_Survived.plot(kind="bar") plt.xticks(ticks=np.arange(7),labels=Sibsp_Survived.index,rotation=360) plt.title("兄弟姐妹配偶数量和生还的关系") plt.show()

Sibsp_Survived.iloc[1:,:].plot(kind="bar") plt.xticks(ticks=np.arange(5),labels=Sibsp_Survived.iloc[1:,:].index,rotation=360) plt.title("兄弟姐妹配偶数量和生还的关系") plt.show()

5、同船家人数和生还的关系

Parch_Survived = pd.crosstab(index=data_train["Parch"],columns=data_train["Survived"]) Parch_Survived.plot(kind="bar") plt.xticks(np.arange(7),Parch_Survived.index,rotation=360) plt.title("同船家人数和生还的关系") plt.show()

Parch_Survived.iloc[3:,:].plot(kind="bar") plt.xticks(np.arange(4),Parch_Survived.iloc[3:,:].index,rotation=360) plt.title("同船家人数和生还的关系") plt.show()

6、船票是否共有与生还的关系

data_train["Ticket"].unique().size

681,小于船客总数，说明有共用票，新建特征，将共用票的置为1，反正，为0

不将分组类别作为索引（as_index = False） 统计不同Ticket使用者数量

Ticket_Count = data_train.groupby("Ticket",as_index = False)["PassengerId"].count() 结果： Ticket PassengerId 0 110152 3 1 110413 3 2 110465 2 3 110564 1 4 110813 1 ... ... ... 676 W./C. 6608 4 677 W./C. 6609 1 678 W.E.P. 5734 1 679 W/C 14208 1 680 WE/P 5735 2 681 rows × 2 columns

提取使用者为1的Ticket id，将data_train中满足Ticket使用者为1的置成0，其余的置成1

Ticket_Count_0 = Ticket_Count[Ticket_Count.PassengerId == 1]['Ticket'] data_train['GroupTicket'] = np.where(data_train.Ticket.isin(Ticket_Count_0), 0, 1) GroupTicket_Survived = pd.crosstab(data_train["GroupTicket"],data_train["Survived"]) GroupTicket_Survived.plot(kind="bar") plt.xticks(np.arange(2),GroupTicket_Survived.index,rotation=360) plt.show()

7、船票价格和生还的关系

bins = np.arange(-50,600,50) data_train["Fare_range"] = pd.cut(data_train["Fare"],bins=bins) Fare_range_Survived = pd.crosstab(data_train["Fare_range"],data_train["Survived"]) Fare_range_Survived.plot(kind="bar") plt.xticks(np.arange(8),Fare_range_Survived.index) plt.show()

Fare_range_Survived.iloc[2:,:].plot(kind="bar") plt.xticks(np.arange(6),Fare_range_Survived.iloc[2:,:].index,rotation=360) plt.show()

二、 处理缺失值之前保证数据完整性，copy一份

train = data_train.copy()

1、"Cabin"使用No填充,表示没座位

train["Cabin"].fillna("No",inplace=True) Cabin_Survived = pd.crosstab(train["Cabin"],train["Survived"]) train["Group_Cabin"] = np.where(train["Cabin"]=="No",0,1) Group_Cabin_Survived = pd.crosstab(train["Group_Cabin"],train["Survived"]) Group_Cabin_Survived.plot(kind="bar") plt.xticks(np.arange(2),Group_Cabin_Survived.index,rotation=360) plt.show()

2、"Embarked"缺失两个使用众数填充

train["Embarked"].fillna(train["Embarked"].mode()[0],inplace=True) train["Embarked_New"] = train["Embarked"].map({"S":0,"C":1,"Q":2}) Embarked_New_Survived = pd.crosstab(train["Embarked_New"],train["Survived"]) Embarked_New_Survived.plot(kind="bar") plt.xticks(np.arange(3),Embarked_New_Survived.index) plt.show()

3、"Age"采用头衔对应的中位数填充

Age_Appellation_median = train.groupby('Appellation')['Age'].median() train.set_index(keys=train["Appellation"],inplace=True) train["Age"].fillna(Age_Appellation_median,inplace=True)

Age缺失值填充完成后，使用reset_index，其中，drop表示不保留以前的索引，也就是"Appellation"，该特征在训练集后边仍有，可不保存，否则因为冲突报错

train = train.reset_index(drop=True) bins = np.arange(0,90,10) train["Age_Group"] = pd.cut(train["Age"], bins=bins) Age_Group_Survived = pd.crosstab(train["Age_Group"],train["Survived"]) Age_Group_Survived.plot(kind="bar") plt.xticks(np.arange(8),Age_Group_Survived.index,rotation=360) plt.show()

三、新特征创建

train["Sex"] = train["Sex"].map({"male":0,"female":1}) dict = DictVectorizer(sparse=False) Appellation_new = dict.fit_transform(pd.DataFrame(train["Appellation"]).to_dict(orient="records")) Appellation_new = pd.DataFrame(Appellation_new) train = pd.merge(train,Appellation_new,left_index=True,right_index=True) train["Age_reserved"] = train["Age"] train.loc[(train["Age"]>0)&(train["Age"]<=10),"Age"] =0 train.loc[(train["Age"]>10)&(train["Age"]<=20),"Age"] =1 train.loc[(train["Age"]>20)&(train["Age"]<=30),"Age"] =2 train.loc[(train["Age"]>30)&(train["Age"]<=40),"Age"] =3 train.loc[(train["Age"]>40)&(train["Age"]<=50),"Age"] =4 train.loc[(train["Age"]>50)&(train["Age"]<=60),"Age"] =5 train.loc[(train["Age"]>60)&(train["Age"]<=70),"Age"] =6 train.loc[(train["Age"]>70)&(train["Age"]<=80),"Age"] =7 train["Fare_reserved"] = train["Fare"] train.loc[(train["Fare"]>-50)&(train["Fare"]<=0),"Fare"] = 0 train.loc[(train["Fare"]>0)&(train["Fare"]<=50),"Fare"] = 1 train.loc[(train["Fare"]>50)&(train["Fare"]<=100),"Fare"] = 2 train.loc[(train["Fare"]>100)&(train["Fare"]<=150),"Fare"] = 3 train.loc[(train["Fare"]>150)&(train["Fare"]<=200),"Fare"] = 4 train.loc[(train["Fare"]>200)&(train["Fare"]<=250),"Fare"] = 5 train.loc[(train["Fare"]>250)&(train["Fare"]<=300),"Fare"] = 6 train.loc[(train["Fare"]>300)&(train["Fare"]<=350),"Fare"] = 7 train.loc[(train["Fare"]>350)&(train["Fare"]<=400),"Fare"] = 8 train.loc[(train["Fare"]>400)&(train["Fare"]<=450),"Fare"] = 9 train.loc[(train["Fare"]>450)&(train["Fare"]<=500),"Fare"] = 10 train.loc[(train["Fare"]>500)&(train["Fare"]<=550),"Fare"] = 11 train["Family"] = train["SibSp"] + train["Parch"] + 1 train.drop(["PassengerId","Name","Ticket","Cabin","Embarked","Appellation","Fare_range","Age_Group"],axis=1,inplace=True)

四、模型搭建

x = train.drop("Survived",axis=1).values y = train["Survived"].values x_train,x_test,y_train,y_test = train_test_split(x,y,test_size=0.25,random_state=0) print(x_train.shape) print(y_train.shape) print(x_test.shape) print(y_test.shape) 结果： (668, 17) (668,) (223, 17) (223,) knn = KNeighborsClassifier() dec = DecisionTreeClassifier() rf = RandomForestClassifier() param = {"n_neighbors":[2,3,4,5,6,7,8,9,10]} grid = GridSearchCV(estimator=knn,param_grid=param,cv=5) grid.fit(x_train,y_train) print(grid.best_params_) print(grid.score(x_test,y_test)) 结果： {'n_neighbors': 5} 0.7399103139013453 param = {"max_depth":[2,3,4,5,6]} grid = GridSearchCV(estimator=dec,param_grid=param,cv=5) grid.fit(x_train,y_train) print(grid.best_params_) print(grid.score(x_test,y_test)) 结果： {'max_depth': 3} 0.7982062780269058 param = {"max_depth":[2,3,4,5,6], "n_estimators":[10,30,50,70,100,120,150,170,200]} grid = GridSearchCV(estimator=rf,param_grid=param,cv=5) grid.fit(x_train,y_train) print(grid.best_params_) print(grid.score(x_test,y_test)) 结果： {'max_depth': 6, 'n_estimators': 120} 0.820627802690583 logis = LogisticRegression() logis.get_params param = {"C":[0.01,0.03,0.1,0.3,1,2], "penalty":["l1","l2"]} grid = GridSearchCV(estimator=logis,param_grid=param,cv=5) grid.fit(x_train,y_train) print(grid.best_params_) print(grid.score(x_test,y_test)) 结果： {'C': 1, 'penalty': 'l1'} 0.8295964125560538 data_test["Appellation"] = data_test["Name"].apply(lambda x : re.search("\w+\.",x).group()).str.replace(".","") data_test['Appellation'].replace(['Capt','Col','Countess','Don','Dr','Jonkheer','Lady','Major','Rev','Sir'], 'Rare',inplace=True) data_test['Appellation'].replace(['Mlle','Ms'], 'Miss',inplace=True) data_test['Appellation'].replace('Mme', 'Mrs',inplace=True)

五、对测试集进行同样特征处理操作

data_test["Appellation"] = data_test["Name"].apply(lambda x : re.search("\w+\.",x).group()).str.replace(".","") data_test['Appellation'].replace(['Capt','Col','Countess','Don','Dr','Jonkheer','Lady','Major','Rev','Sir'], 'Rare',inplace=True) data_test['Appellation'].replace(['Mlle','Ms'], 'Miss',inplace=True) data_test['Appellation'].replace('Mme', 'Mrs',inplace=True) test = data_test.copy() Appellation_new_test = pd.DataFrame(dict.transform(pd.DataFrame(test["Appellation"]).to_dict(orient="records"))) test = pd.merge(test,Appellation_new_test,left_index=True,right_index=True) test["Cabin"].fillna("No",inplace=True) test["Group_Cabin"] = np.where(test["Cabin"]=="No",0,1) test["Embarked_New"] = test["Embarked"].map({"S":0,"C":1,"Q":2}) test["Sex"] = test["Sex"].map({"male":0,"female":1}) test_median = test.groupby("Appellation")["Age"].median() test.set_index("Appellation",inplace=True) test["Age"].fillna(test_median ,inplace=True) test.reset_index(drop=False,inplace=True) bins = np.arange(0,90,10) test["Age_range"] = pd.cut(test["Age"],bins=bins) test["Age_range"].value_counts(dropna=False) test["Age_reserved"] = test["Age"] test.loc[(test["Age"]>0)&(test["Age"]<=10),"Age"] =0 test.loc[(test["Age"]>10)&(test["Age"]<=20),"Age"] =1 test.loc[(test["Age"]>20)&(test["Age"]<=30),"Age"] =2 test.loc[(test["Age"]>30)&(test["Age"]<=40),"Age"] =3 test.loc[(test["Age"]>40)&(test["Age"]<=50),"Age"] =4 test.loc[(test["Age"]>50)&(test["Age"]<=60),"Age"] =5 test.loc[(test["Age"]>60)&(test["Age"]<=70),"Age"] =6 test.loc[(test["Age"]>70)&(test["Age"]<=80),"Age"] =7 test["Fare"].fillna(test["Fare"].mean(),inplace=True) test["Fare_reserved"] = test["Fare"] test.loc[(test["Fare"]>-50)&(test["Fare"]<=0),"Fare"] = 0 test.loc[(test["Fare"]>0)&(test["Fare"]<=50),"Fare"] = 1 test.loc[(test["Fare"]>50)&(test["Fare"]<=100),"Fare"] = 2 test.loc[(test["Fare"]>100)&(test["Fare"]<=150),"Fare"] = 3 test.loc[(test["Fare"]>150)&(test["Fare"]<=200),"Fare"] = 4 test.loc[(test["Fare"]>200)&(test["Fare"]<=250),"Fare"] = 5 test.loc[(test["Fare"]>250)&(test["Fare"]<=300),"Fare"] = 6 test.loc[(test["Fare"]>300)&(test["Fare"]<=350),"Fare"] = 7 test.loc[(test["Fare"]>350)&(test["Fare"]<=400),"Fare"] = 8 test.loc[(test["Fare"]>400)&(test["Fare"]<=450),"Fare"] = 9 test.loc[(test["Fare"]>450)&(test["Fare"]<=500),"Fare"] = 10 test.loc[(test["Fare"]>500)&(test["Fare"]<=550),"Fare"] = 11 test["Family"] = test["SibSp"] + test["Parch"] + 1 test["Ticket"].value_counts() test_count0 = test.groupby("Ticket",as_index=False)["PassengerId"].count() test_count0 = test_count0[test_count0["PassengerId"]==1]["Ticket"] test['GroupTicket'] = np.where(test["Ticket"].isin(test_count0), 0, 1) test.drop(["PassengerId","Name","Ticket","Cabin","Embarked","Appellation","Age_range"],axis=1,inplace=True) order_train = train.drop("Survived",axis=1) order = list(order_train.columns) test = test[order] x_pre = test.values param = {"max_depth":[2,3,4,5,6], "n_estimators":[10,30,50,70,100,120,150,170,200]} grid = GridSearchCV(estimator=rf,param_grid=param,cv=5) grid.fit(x_train,y_train) print(grid.score(x_test,y_test)) y_pre = grid.predict(x_pre) print(grid.best_params_) pd.DataFrame(y_pre).to_csv("pre123.csv",index=False,header=False)