【机器学习kaggle赛事】泰坦尼克号生存预测

写在前面

回顾这个项目的基本流程：

1、查看数据集，合并训练集测试集以一起进行数据清洗

2、数据清洗：查看数据集空缺值，并填充空缺值

3、探索性可视化：通过透视表和图表，探究各个特征与label（需要预测的值）的相关性，若有相关性则保留。

4、特征工程：

若特征为离散型特征（类别型特征）（Name,Pclass,Embarked，Sex等），则进行独热编码，若特征为连续型数值（Fare,Age），则对特征数值进行分组，并对不同组别进行赋值，再进行Ont_hot编码处理
属性几乎相同的多个特征（如SibSp: 乘客在船上的兄妹姐妹数/配偶数，Parch: 乘客在船上的父母数/子女数都属于亲人），在分别探究其与label的相关性决定保留后，可以将它们重组为新的特征。

4、特征选择：通过相关系数来选择

数据集情况查看

将训练数据与测试数据连接起来，以便一起进行数据清洗。

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
 
train = pd.read_csv('train.csv')
test = pd.read_csv('test.csv')
    
    # 这里需要注意的是，如果没有后面的ignore_index=True
    #那幺index的值在连接后的这个新数据中是不连续的 继续从 0开始，如果要按照index删除一行数据，可能会发现多删一条。
full = pd.concat([train, test], ignore_index=True)
full.head()  # 默认显示5行

PassengerId: 乘客ID；
Survived: 生存情况，0代表不幸遇难，1代表存活；
Pclass: 仓位等级，1为一等舱，2为二等舱，3为三等舱；
Name: 乘客姓名；
Sex: 性别；
Age: 年龄；
SibSp: 乘客在船上的兄妹姐妹数/配偶数（即同代直系亲属数）；
Parch: 乘客在船上的父母数/子女数（即不同代直系亲属数）；
Ticket: 船票编号；
Fare: 船票价格；
Cabin: 客舱号；
Embarked: 登船港口（S: Southampton; C: Cherbourg Q: Queenstown）

#查看字符型数据情况：
full.describe(include=['O'])

full.describe().T
#describe()函数只能查看数据类型的描述统计信息，无法查看类似字符类型的信息。
#故需用info()函数进一步查看每一列的数据信息。
full.info()

数据清洗

print(full.isnull().sum()) # 查看数据的缺失情况

PassengerId       0
Survived        418
Pclass            0
Name              0
Sex               0
Age             263
SibSp             0
Parch             0
Ticket            0
Fare              1
Cabin          1014
Embarked          2
dtype: int64

缺失值 Age 263， Fare 1 ，Cabin 1014， Embarked 2

Embarked：

分类数据，使用最常见的类别取代，用众数填充。

#用众数填补Embarked    
#查看众数
full.Embarked.mode()
full['Embarked'].fillna('S',inplace=True)

Fare

船票费用是类别数据，但是不同客舱的票价不同 ，用Fare空缺所在的客舱号的票价中位数来填充

#填补Fare空缺值,用Pclass==3的客舱票价的中位数来填充
full[full.Fare.isnull()]#查看Fare缺失值的信息，获得Pclass
full.Fare.fillna(full[full.Pclass==3]['Fare'].median(),inplace=True)

Age

用中位数来填充年龄

如果是数值类型，使用平均值或者中位数进行填充

#年龄(Age) 最小值为0.17，不存在0值，其数据缺失率为263/1309=20.09%，由于Age的平均数与中位数接近，故选择平均值作为缺失项的填充值。
full.Age.fillna(full.Age.mean(),inplace = True)
# full['Age']=full['Age'].fillna(full['Age'].mean())
full.Age.describe()
'''count    1309.000000
mean       29.881138
std        12.883193
min         0.170000
25%        22.000000
50%        29.881138
75%        35.000000
max        80.000000
Name: Age, dtype: float64'''

Cabin

1,用U来填充，表示unkown

字符串类型，按照实际情况填写，无法追踪的信息，用”Unknow”填充。处理Cabin缺失值 U代表Unknow

 full['Cabin'] = full['Cabin'].fillna('U')

2缺失个数：

687 超过75%的数据缺失，故不打算填补。考虑以Cabin是否缺失来构建一个新特征，看是否对生存有影响。若没有影响，则删除该列

full.Cabin.isnull().sum()
#1014
#创造新特征Cabin_exist，判断Cabin与Survived的相关性，相关性不大则删除
full['Cabin_exist'] = full['Cabin'].map(lambda x:"Yes" if type(x)==str else "No")# 判断类型是否相等
full[["Cabin_exist", "Survived"]].groupby("Cabin_exist",as_index=False).mean()

full =full.drop('Cabin_exist',axis = 1)
full.head()
#将字符转换为数值
full['Cabin_exist'] = full['Cabin'].map(lambda x: 1 if type(x)==str else 0)
full = full.drop('Cabin',axis = 1)
full.head()
# full.loc[full.Cabin.notnull(),'Cabin'] = 1
# full.loc[full.Cabin.isnull(),'Cabin'] = 0
# full.Cabin.isnull().sum()#验证填充效果
sns.barplot(x="Cabin_exist",y="Survived",data=full)
full.isnull().sum()
#数据处理完毕
full.head()

特征工程

1，探究Sex与Survived的相关性

'''======================================================特征工程========================================================================'''
#1，探究Sex与Survived的相关性
full[['Sex','Survived']].groupby('Sex',as_index = False).mean().sort_values('Survived',ascending = False)
sns.countplot(x = 'Sex',hue = 'Survived',data =full)

#Sex（性别）,将字符映射到字符
sex_dict = {'male': 1,'female':0}
full['Sex'] = full['Sex'].map(sex_dict)
full['Sex'].head()
full[["Pclass","Sex","Survived"]].groupby(["Pclass","Sex"],as_index=False).mean().sort_values(by="Survived",ascending=False)
#两维变量关系图
sns.factorplot(x="Pclass",y="Survived",hue="Sex",data=full)

2，探究Pcalss与Survived的关联性

 #2，探究Pcalss与Survived的关联性,相关性较高
full[["Pclass","Survived"]].groupby(["Pclass"],as_index = False).mean().sort_values(by="Survived",ascending=False)
sns.barplot(x="Pclass",y="Survived",data=full)

可以看出，相关性较高，保留特征Pclass

#将客舱类型进行独热编码
PclassDf = pd.get_dummies(full['Pclass'],prefix = 'Pclass')
PclassDf.head()
#将编码后的客舱特征与原数据合并
full = pd.concat([full,PclassDf],axis = 1)
full.head()
#删除Pclass
ful = full.drop('Pclass',axis = 1)
full.head()

3，Embarked：不同的上船地点对生存率是否有影响

full[["Embarked","Survived"]].groupby("Embarked",as_index=False).count().sort_values("Survived",ascending=False)
sns.barplot(x="Embarked",y="Survived",data=full)

sns.factorplot(x="Sex",y="Survived",hue="Embarked",data=full)
full[["Sex","Survived","Embarked"]].groupby(["Sex","Embarked"],as_index=False).count().sort_values("Survived",ascending=False)
"""S口岸，登船人数644，女性乘客占比46%；C口岸，登船人数168，女性占比接近77%；
Q口岸，登船人数77，女性占比接近88%。
前面已知女性生存率明显高于男性生存率，所以上述问题可能由性别因素引起。"""

#对Embarked进行独热编码
EmbarkedDf = pd.get_dummies(full['Embarked'],prefix = 'Embarked')
EmbarkedDf.head()
# 将EmbarkedDf的特征添加至full数据集
full = pd.concat([full,EmbarkedDf],axis = 1)#按列插入数据
full.head()
# 因为已经使用登船港口(Embarked)进行了one-hot编码产生了它的虚拟变量（dummy variables）
# 所以这里把登船港口(Embarked)删掉
full = full.drop('Embarked',axis = 1)
full.head()

4，Name与Survived的相关性

从name里提取title(头衔)，地位高的人在灾难里会更有可能获得救援从而得以存活

def getTitle(Name):
    s1 = Name.split(',')[1]#Braund, Mr. Owen Harris分割完取第二个元素Mr. Owen Harris
    s2 = s1.split('.')[0]#得到Mr
    return s2.strip( ) #strip() 方法用于移除字符串头尾指定的字符（这里是空格）
full['Title'] = full['Name'].map(getTitle)#将getTitle映射给Name  得到Title
full['Title'].value_counts()
full.drop('Name',axis = 1,inplace  = True)
full.head()

将头衔分为五类：

Officer：政府官员；
Royalty：王室（皇室）；
Mr：已婚男士；
Mrs：已婚妇女；
Miss：年轻未婚女子；
Master：有技能的人/教师

title_dict={"Capt":"Officer","Col":"Officer","Major":"Officer","Jonkheer":"Royalty","Don":"Royalty","Sir":"Royalty","Dr":"Officer","Rev":"Officer"
            ,"the Countess":"Royalty","Dona":"Royalty","Mme":"Mrs","Mlle":"Miss","Ms":"Mrs","Mr" :"Mr","Mrs" :"Mrs","Miss" :"Miss"
            ,"Master" :"Master", "Lady" : "Royalty"}
full['Title']=full['Title'].map(title_dict)#将title字典映射给name
full.head()

title跟Sex有联系，联合起来分析

pd.crosstab(full.Title,full.Sex)#透视表
#探索title与生存的关系
full[["Title","Survived"]].groupby("Title",as_index=False).mean().sort_values("Survived",ascending=False)
sns.barplot(x="Title",y="Survived",data=full)

full['Title'].value_counts()

Mr         757
Miss       262
Mrs        200
Master      61
Officer     23
Royalty      6
Name: Title, dtype: int64

#独热编码
TitleDf = pd.get_dummies(full['Title'],prefix = 'Title')
#添加进full
full = pd.concat([full,TitleDf],axis = 1)
full.head()
#删除不需要的列
full = full.drop(['Title'],axis = 1)
full.head()

5，Cabin与Survived之间的相关性

#客场号的类别值是首字母，因此我们提取客舱号的首字母为特征

# full[‘Cabin’] = full[‘Cabin’].map(lambda x : x[0])

# full[‘Cabin’].value_counts

# full[‘Cabin’]=full[‘Cabin’].map(lambda x:x[0])

# full[‘Cabin’].value_counts()

“””因为之前用处理过，所以不再提取首字母，如果用u来填充的话，可以用这种方法”””

6,探究孤身一人和有家人陪伴的生存率（SibSp，Parch）

full[['SibSp','Survived']].groupby(['SibSp'],as_index = False).mean().sort_values('Survived',ascending = False)
sns.barplot(x = 'SibSp',y = 'Survived',data = full)

full[["Parch","Survived"]].groupby("Parch",as_index=False).mean().sort_values("Survived",ascending=False)
sns.barplot(x = 'Parch',y = 'Survived',data = full)

可以看出SibSp，Parch 跟Survived 有相关联系，所以用SibSp，Parch构建家庭人数和家庭类别的新特征Alone，family_small，family_big

# 构建家庭人数和家庭类别的新特征
full['family'] = full['Parch'] + full['SibSp'] + 1#Parch，SibSp为0时，只有自己一个人，+1
full['Alone'] = np.where(full['family'] == 1,1,0)
full['family_small'] = np.where((full['family']>=2) & (full['family']<=4),1,0)
full['family_big'] = np.where(full['family']>=5,1,0)
full.head()

探究家庭与存活率的关系  证实孤身一人存活率不高

#探究家庭与存活率的关系  证实孤身一人存活率不高
full[['family','Survived']].groupby('family',as_index = False).mean().sort_values('Survived',ascending = False)
sns.barplot(x = 'family',y = 'Survived',data = full)
#孤身一人对生存率是否有影响
full[["Alone","Survived"]].groupby("Alone",as_index = False).mean().sort_values("Survived",ascending=False)
sns.barplot(x="Alone",y="Survived",data=full)
full = full.drop('family',axis = 1)

sns.factorplot(x="Pclass",y="Survived",hue="Alone",data=full)

7、探究年龄Age与Survived的相关性

#查看Age的分布情况
sns.violinplot(y="Age",data=full)
#查看生存与死亡乘客的年龄分布
sns.violinplot(y="Age",x="Survived",data=full)
#将年龄分为五组
full['AgeCut'] = pd.cut(full.Age,5)#要用full['AgeCut']，而不是full.AgeCut,这样AgeCut才会在Index里
full.AgeCut.value_counts().sort_index()
full[['AgeCut','Survived']].groupby('AgeCut',as_index = False).mean().sort_values('Survived',ascending = False)

(0.0902, 16.136]    134
(16.136, 32.102]    787
(32.102, 48.068]    269
(48.068, 64.034]    106
(64.034, 80.0]       13
Name: AgeCut, dtype: int64

根据各个分段，重新给Age赋值并进行独热编码

#根据各个分段，重新给Age赋值
full.loc[full.Age <= 16.136,'Age'] = 1
full.loc[(full.Age > 16.136)&(full.Age <=32.102),'Age'] = 2
full.loc[(full.Age > 32.102)&(full.Age <=48.068),'Age'] = 3
full.loc[(full.Age > 48.068)&(full.Age <=64.034),'Age'] = 4
full.loc[full.Age > 64.034,'Age'] = 5
full.head()
AgeDf = pd.get_dummies(full['Age'],prefix =' Age')
full = pd.concat([full,AgeDf],axis = 1)
full = full.drop(['Age','AgeCut'],axis = 1)
full.head()

8、Fare与Survived之间的相关性

sns.violinplot(y="Fare",data=train)
#对比生死乘客的票价
sns.violinplot(y="Fare",x="Survived",data=train)

# 当然这里也可以用seaborn的displot进行绘制，但是displot的纵坐标是比率，hist的纵坐标是实际个数count；
# figsize调整画布大小
full['Fare'].hist(color='green', bins=30, figsize=(8,4))

#分组
full['FareCut'] = pd.cut(full.Fare,5)
full.FareCut.value_counts().sort_index()
# full.head()
full[['FareCut','Survived']].groupby('FareCut',as_index = False).mean().sort_values('Survived',ascending = False)
#重新赋值
full.loc[full.Fare<=7.854,'Fare']=1
full.loc[(full.Fare>7.854)&(full.Fare<=10.5),'Fare']=2
full.loc[(full.Fare>10.5)&(full.Fare<=21.558),'Fare']=3
full.loc[(full.Fare>21.558)&(full.Fare<=41.579),'Fare']=4
full.loc[full.Fare>41.579,'Fare']=5
full.head()
#进行独热编码
FareDf = pd.get_dummies(full['Fare'],prefix = 'Fare')
full = pd.concat([full,FareDf],axis = 1)
full = full.drop(['Fare','FareCut'],axis = 1)
full.head()
full = full.drop(['SibSp','Parch','Pclass'],axis = 1)
full.head()

特征选择

"""=============================================特征选择和特征降维========================================================="""
full.info()
corr_df=full.corr()  
corr_df
#用图形直观查看线性相关系数
plt.figure(figsize=(16,16))
plt.title("Pearson Correlation of Features")
sns.heatmap(corr_df,linewidths=0.1,square=True,linecolor="white",annot=True,cmap='YlGnBu',vmin=-1,vmax=1)
corr_df['Survived'].sort_values(ascending = False)

Survived         1.000000
Title_Mrs        0.344935
Title_Miss       0.332795
Cabin_exist      0.316912
Pclass_1         0.285904
family_small     0.279855
Fare_5.0         0.266217
Embarked_C       0.168240
 Age_1.0         0.121485
Pclass_2         0.093349
Title_Master     0.085221
Fare_4.0         0.058052
Fare_3.0         0.043153
Title_Royalty    0.033391
 Age_4.0         0.030350
 Age_3.0         0.021711
Embarked_Q       0.003650
Title_Officer   -0.031316
 Age_5.0        -0.067344
 Age_2.0        -0.097245
family_big      -0.125147
Embarked_S      -0.149683
Fare_1.0        -0.164287
Fare_2.0        -0.198067
Alone           -0.203367
Pclass_3        -0.322308
Sex             -0.543351
Title_Mr        -0.549199
Name: Survived, dtype: float64

构建包含最终选择特征的数据集

最终选择特征如下：

full_x=pd.concat([TitleDf,PclassDf,EmbarkedDf,FareDf,AgeDf,full['Sex'],full['family_small'],full['Cabin_exist']
                  ,full['family_big'],full['Alone']],axis=1)
full_x.head()

分割训练数据和测试数据

​
#前891行为原始训练数据
source_x=full_x.loc[0:890,:]#提取特征值
source_y=full.loc[0:890,'Survived']#提取标签值
#后418行是test,预测数据
pred_x=full_x.loc[891:,:]
source_x.shape#(891, 23)
source_y.shape#(891,)
pred_x.shape#(418, 23)
​

构建训练集和数据集

#训练数据集和测试数据集，按照二八原则分为训练数据和测试数据，其中80%为训练数据
from sklearn.model_selection import train_test_split
x_train,x_test,y_train,y_test =train_test_split(source_x,source_y,train_size=0.8)
print('训练数据集特征:{0},训练数据集标签:{1}'.format(x_train.shape,y_train.shape))
print('测试数据集特征:{0},测试数据集标签:{1}'.format(x_test.shape,y_test.shape))
#训练数据集特征:(712, 27),训练数据集标签:(712,)
#测试数据集特征:(179, 27),测试数据集标签:(179,)

#对x_train,x_test进行标准化
from sklearn.preprocessing import StandardScaler
sc = StandardScaler()
x_train_std=sc.fit_transform(x_train)
x_test_std=sc.transform(x_test)

不同模型对比

from sklearn.model_selection import cross_val_score
from sklearn.neighbors import KNeighborsClassifier
from sklearn.naive_bayes import GaussianNB
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.svm import SVC
models=[KNeighborsClassifier(),GaussianNB(),DecisionTreeClassifier(),RandomForestClassifier(),
       GradientBoostingClassifier(),SVC()]
# 计算各模型得分
names=['KNN','NB','Tree','RF','GDBT','SVM']
for name, model in zip(names,models):
    score=cross_val_score(model,Xa,y,cv=5)
    print("{}:{},{}".format(name,score.mean(),score))

KNN:0.8159374803841566,[0.81564246 0.76404494 0.84269663 0.81460674 0.84269663]
NB:0.7385412089636557,[0.69832402 0.74719101 0.82022472 0.57303371 0.85393258]
Tree:0.8069612704789405,[0.80446927 0.76404494 0.85393258 0.79775281 0.81460674]
RF:0.8125666938673028,[0.81564246 0.75842697 0.8258427  0.82022472 0.84269663]
GDBT:0.8092147385600402,[0.79888268 0.7752809  0.83146067 0.78651685 0.85393258]
SVM:0.821542903772519,[0.82681564 0.80337079 0.8258427  0.78651685 0.86516854]

# 使用标准化的数据 scaled data
names=['KNN','NB','Tree','RF','GDBT','SVM']
for name, model in zip(names,models):
    score=cross_val_score(model,X_std,y,cv=5)
    print("{}:{},{}".format(name,score.mean(),score))

KNN:0.8204569706860838,[0.79329609 0.76404494 0.87640449 0.82022472 0.84831461]
NB:0.7216872763793861,[0.69832402 0.74719101 0.82022472 0.48876404 0.85393258]
Tree:0.8092084614901763,[0.80446927 0.76404494 0.84831461 0.79775281 0.83146067]
RF:0.8226727763480006,[0.82122905 0.76966292 0.85955056 0.81460674 0.84831461]
GDBT:0.8103383340656581,[0.79888268 0.7752809  0.83146067 0.78651685 0.85955056]
SVM:0.8170610758897746,[0.81564246 0.80337079 0.8258427  0.78089888 0.85955056]

通过决策树看各个特征的重要性

clf = DecisionTreeClassifier(criterion = 'entropy',random_state = 30,splitter = 'random')
clf.fit(x_train_std,y_train)
score = clf.score(x_test_std,y_test)
score
fi=pd.DataFrame({'importance':clf.feature_importances_},index=X.columns)
fi.sort_values('importance',ascending=False)

fi.sort_values('importance',ascending=False).plot.bar(figsize=(11,7))
plt.xticks(rotation=30)
plt.title('Feature Importance',size='x-large')

from sklearn.model_selection import GridSearchCV
param_grid={'n_neighbors':[1,2,3,4,5,6,7,8,9]}
grid_search=GridSearchCV(KNeighborsClassifier(),param_grid,cv=5)
 
grid_search.fit(x_train_std,y_train)
 
grid_search.best_params_,grid_search.best_score_

#LogisticRegression
param_grid={'C':[0.01,0.1,1,10]}
grid_search=GridSearchCV(LogisticRegression(),param_grid,cv=5)
 
grid_search.fit(x_train_std,y_train)
 
grid_search.best_params_,grid_search.best_score_
# second round grid search
param_grid={'C':[0.04,0.06,0.08,0.1,0.12,0.14]}
grid_search=GridSearchCV(LogisticRegression(),param_grid,cv=5)
 
grid_search.fit(x_train_std,y_train)
 
grid_search.best_params_,grid_search.best_score_

#Support Vector Machine
param_grid={'C':[0.01,0.1,1,10],'gamma':[0.01,0.1,1,10]}
grid_search=GridSearchCV(SVC(),param_grid,cv=5)
 
grid_search.fit(x_train_std,y_train)
 
grid_search.best_params_,grid_search.best_score_
 
#second round grid search
param_grid={'C':[2,4,6,8,10,12,14],'gamma':[0.008,0.01,0.012,0.015,0.02]}
grid_search=GridSearchCV(SVC(),param_grid,cv=5)
 
grid_search.fit(x_train_std,y_train)
 
grid_search.best_params_,grid_search.best_score_

#Gradient Boosting Decision Tree
param_grid={'n_estimators':[30,50,80,120,200],'learning_rate':[0.05,0.1,0.5,1],'max_depth':[1,2,3,4,5]}
grid_search=GridSearchCV(GradientBoostingClassifier(),param_grid,cv=5)
 
grid_search.fit(x_train_std,y_train)
 
grid_search.best_params_,grid_search.best_score_
#second round grid search
param_grid={'C':[2,4,6,8,10,12,14],'gamma':[0.008,0.01,0.012,0.015,0.02]}
grid_search=GridSearchCV(SVC(),param_grid,cv=5)
 
grid_search.fit(x_train_std,y_train)
 
grid_search.best_params_,grid_search.best_score_

#Gradient Boosting Decision Tree
param_grid={'n_estimators':[30,50,80,120,200],'learning_rate':[0.05,0.1,0.5,1],'max_depth':[1,2,3,4,5]}
grid_search=GridSearchCV(GradientBoostingClassifier(),param_grid,cv=5)
 
grid_search.fit(x_train_std,y_train)
 
grid_search.best_params_,grid_search.best_score_
#second round search
param_grid={'n_estimators':[100,120,140,160],'learning_rate':[0.05,0.08,0.1,0.12],'max_depth':[3,4]}
grid_search=GridSearchCV(GradientBoostingClassifier(),param_grid,cv=5)
 
grid_search.fit(x_train_std,y_train)
 
grid_search.best_params_,grid_search.best_score_

模型训练

#逻辑回归
from sklearn.linear_model import LogisticRegression
model1 = LogisticRegression()
model1.fit(x_train_std,y_train)
LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True,
              intercept_scaling=1, max_iter=100, multi_class='ovr', n_jobs=1,
              penalty='l2', random_state=None, solver='liblinear', tol=0.0001,
              verbose=0, warm_start=False)
#在测试集上得出模型正确率
model1.score(x_test_std,y_test)
0.8212290502793296
pred1 = model1.predict(x_test)
score1 = model1.score(x_train,y_train)
score1
#使用训练得到的模型对pred_x的生存情况进行预测
pred_x_std=sc.fit_transform(pred_x)
pred_y=model1.predict(pred_x_std)
pred_y