python泰坦尼克号存活率分析

对泰坦尼克号的数据 进行特征分析、数据清理、数据填充、处理分类特征、将连续特征转化为离散特征、合并特征、制作模型、模型预测

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
#将警告不显示
import warnings
warnings.filterwarnings(&＃39;ignore&＃39;)#导入测试集和训练集
train_df&＃61;pd.read_csv(&＃39;train.csv&＃39;)
# train_df.head()
test_df&＃61;pd.read_csv(&＃39;test.csv&＃39;)
combine&＃61;[train_df,test_df]PassengerId&＃61;test_df[&＃39;PassengerId&＃39;]#因为后续会将该列删除 所以先对这一列数据进行备份


特征分析
分析存活率与几个因素之间的关系

#pclass与survived的关系
train_df[[&＃39;Pclass&＃39;,&＃39;Survived&＃39;]].groupby([&＃39;Pclass&＃39;],as_index&＃61;False).mean().sort_values(by&＃61;&＃39;Survived&＃39;,ascending&＃61;False)


#Sex与survived的关系
train_df[[&＃39;Sex&＃39;,&＃39;Survived&＃39;]].groupby([&＃39;Sex&＃39;],as_index&＃61;False).mean().sort_values(by&＃61;&＃39;Survived&＃39;,ascending&＃61;False)


#SibSp与Survived的关系
train_df[[&＃39;SibSp&＃39;,&＃39;Survived&＃39;]].groupby([&＃39;SibSp&＃39;],as_index&＃61;False).mean().sort_values(by&＃61;&＃39;Survived&＃39;,ascending&＃61;False)


#Parch与Survived的关系
train_df[[&＃39;Parch&＃39;,&＃39;Survived&＃39;]].groupby([&＃39;Parch&＃39;],as_index&＃61;False).mean().sort_values(by&＃61;&＃39;Survived&＃39;,ascending&＃61;False)


#Embarked与Survived的关系
train_df[[&＃39;Embarked&＃39;,&＃39;Survived&＃39;]].groupby([&＃39;Embarked&＃39;],as_index&＃61;False).mean().sort_values(by&＃61;&＃39;Survived&＃39;,ascending&＃61;False)


#绘制年龄与死亡率、存活率的条形图
fig&＃61;plt.figure(figsize&＃61;(12,8))ax1&＃61;fig.add_subplot(1,2,1)
ax2&＃61;fig.add_subplot(1,2,2)age1&＃61;train_df.Age[train_df[&＃39;Survived&＃39;]&＃61;&＃61;0]
age2&＃61;train_df.Age[train_df[&＃39;Survived&＃39;]&＃61;&＃61;1]ax1.set_title(&＃39;Survived&＃61;0&＃39;)
ax1.hist(age1,bins&＃61;20)
ax1.set_xlabel(&＃39;age&＃39;)
ax1.set_yticks([10,20,30,40,50,60])ax2.set_title(&＃39;Survived&＃61;1&＃39;)
ax2.hist(age2,bins&＃61;20)
ax2.set_xlabel(&＃39;age&＃39;)
ax2.set_yticks([10,20,30,40,50,60])
#定义同一规格的y轴 方便查看死亡与存货的区别plt.show()


#fare和Survived 绘制图像
fare&＃61;train_df.Fare[train_df[&＃39;Survived&＃39;]&＃61;&＃61;1]
plt.hist(fare)
plt.show()


数据清理

#删除无用特征
train_df&＃61;train_df.drop([&＃39;Ticket&＃39;,&＃39;Cabin&＃39;],axis&＃61;1)
test_df&＃61;test_df.drop([&＃39;Ticket&＃39;,&＃39;Cabin&＃39;],axis&＃61;1)
combine&＃61;[train_df,test_df]
train_df.head()


#创建新特征train_df.Name.str.extract(&＃39; ([a-zA-Z]&＃43;)\.&＃39;,expand&＃61;False)
#通过正则表达式 将姓名中与性别有关的称呼截取出来 创建为新的特征for database in combine:database[&＃39;Title&＃39;]&＃61;database.Name.str.extract(&＃39; ([a-zA-Z]&＃43;)\.&＃39;,expand&＃61;False)pd.crosstab(train_df[&＃39;Title&＃39;],train_df[&＃39;Sex&＃39;])
#crosstab 交叉提取列 两个参数&＃xff08;列&＃xff0c;行&＃xff09;


#分类title 归为几个大类
for database in combine:database[&＃39;Title&＃39;]&＃61;database[&＃39;Title&＃39;].replace([&＃39;Lady&＃39;,&＃39;Countess&＃39;,&＃39;Capt&＃39;,&＃39;Col&＃39;,&＃39;Don&＃39;,&＃39;Dr&＃39;,&＃39;Major&＃39;,&＃39;Rev&＃39;,&＃39;Sir&＃39;,&＃39;Jonkheer&＃39;,&＃39;Dona&＃39;],&＃39;Rare&＃39;)#将出现次数少的归类为Rare一类database[&＃39;Title&＃39;]&＃61;database[&＃39;Title&＃39;].replace(&＃39;Mlle&＃39;,&＃39;Miss&＃39;)#将表达为同一意思的称呼归为一类database[&＃39;Title&＃39;]&＃61;database[&＃39;Title&＃39;].replace(&＃39;Ms&＃39;,&＃39;Miss&＃39;)database[&＃39;Title&＃39;]&＃61;database[&＃39;Title&＃39;].replace(&＃39;Mme&＃39;,&＃39;Mrs&＃39;)
#replace函数 两个参数&＃xff08;被替换内容&＃xff0c;替换后内容&＃xff09; train_df[[&＃39;Title&＃39;,&＃39;Survived&＃39;]].groupby([&＃39;Title&＃39;],as_index&＃61;False).mean().sort_values(by&＃61;&＃39;Survived&＃39;,ascending&＃61;False)


#删除无用两列数据
train_df.drop([&＃39;PassengerId&＃39;,&＃39;Name&＃39;],axis&＃61;1,inplace&＃61;True)
test_df.drop([&＃39;PassengerId&＃39;,&＃39;Name&＃39;],axis&＃61;1,inplace&＃61;True)
combine&＃61;[train_df,test_df]
train_df


数据填充

#快速查找缺失值
train_df.info()


从图中会发现 Age 和 Embarked 都有缺失值

#填充Embarked特征&＃xff08;缺失值较少 利用众数进行填充&＃xff09;
freq&＃61;train_df.Embarked.dropna().mode()[0]
#mode 显示数量最多的数for database in combine:database.Embarked&＃61;database.Embarked.fillna(freq)


#填充Age&＃xff08;缺失值较多 不可以用众数填充 需通过其他特征填充值&＃xff09;
grp &＃61; train_df.groupby([&＃39;Pclass&＃39;,&＃39;Sex&＃39;,&＃39;Title&＃39;])[&＃39;Age&＃39;].mean().reset_index()#定义一个新函数 填充缺失值
def fill_age(x):return grp[(grp[&＃39;Pclass&＃39;]&＃61;&＃61;x[&＃39;Pclass&＃39;]) & (grp[&＃39;Sex&＃39;]&＃61;&＃61;x[&＃39;Sex&＃39;]) & (grp[&＃39;Title&＃39;]&＃61;&＃61;x[&＃39;Title&＃39;])].Age.values[0]#如果缺失值的三个条件满足平均的三个条件&＃xff0c;即填充对应的Age值#用lambda 定义另一个简单的新的函数 判断Age值是否缺失
train_df[&＃39;Age&＃39;] &＃61; train_df.apply(lambda x: fill_age(x) if np.isnan(x[&＃39;Age&＃39;]) else x[&＃39;Age&＃39;] ,axis&＃61;1)
test_df[&＃39;Age&＃39;] &＃61; test_df.apply(lambda x: fill_age(x) if np.isnan(x[&＃39;Age&＃39;]) else x[&＃39;Age&＃39;] ,axis&＃61;1)
combine &＃61; [train_df,test_df]


处理分类特征
将数据根据分类转化为数字 0.1.2.3…

#处理sex 将数据转化为1和0
for database in combine:database[&＃39;Sex&＃39;] &＃61; database[&＃39;Sex&＃39;].map({&＃39;female&＃39;:1,&＃39;male&＃39;:0}).astype(int)


#处理Title
title_mapping &＃61; {&＃39;Mr&＃39;:1 ,&＃39;Miss&＃39;:2 ,&＃39;Mrs&＃39;:3 ,&＃39;Master&＃39;:4 ,&＃39;Rare&＃39;:5}
for database in combine:database[&＃39;Title&＃39;] &＃61; database[&＃39;Title&＃39;].map(title_mapping).astype(int)


将连续特征转化为离散特征
将连续的数据 分为几组\几类

#处理Age离散化#将age分为五组 准确分组
train_df[&＃39;AgeBand&＃39;] &＃61;pd.cut(train_df[&＃39;Age&＃39;],5)
train_df[&＃39;AgeBand&＃39;]


通过切割数据 可以将Age准确的分为五个组 并将不同组用1.2.3.4.5代替

for database in combine:database.loc[ database[&＃39;Age&＃39;] <16,&＃39;Age&＃39;]&＃61;0database.loc[(database[&＃39;Age&＃39;] >&＃61;16)&(database[&＃39;Age&＃39;] <32),&＃39;Age&＃39;]&＃61;1database.loc[(database[&＃39;Age&＃39;] >&＃61;32)&(database[&＃39;Age&＃39;] <48),&＃39;Age&＃39;]&＃61;2database.loc[(database[&＃39;Age&＃39;] >&＃61;48)&(database[&＃39;Age&＃39;] <64),&＃39;Age&＃39;]&＃61;3database.loc[ database[&＃39;Age&＃39;] >&＃61;64,&＃39;Age&＃39;]&＃61;4train_df.drop([&＃39;AgeBand&＃39;],axis&＃61;1,inplace&＃61;True)
combine&＃61;[train_df,test_df]
train_df


利用qcut处理Fare的数据

#处理Fare &＃xff08;不可用cut均分的方法 使用qcut 根据数量均等分&＃xff09;
train_df[&＃39;FareBand&＃39;]&＃61;pd.qcut(train_df[&＃39;Fare&＃39;],4,duplicates&＃61;&＃39;drop&＃39;)#处理test_df中的缺失值&＃xff08;因为缺失1个&＃xff0c;所以选择平均值填充&＃xff09;
test_df[&＃39;Fare&＃39;] &＃61; test_df[&＃39;Fare&＃39;].fillna(test_df[&＃39;Fare&＃39;].mean())for dataset in combine:dataset.loc[ dataset[&＃39;Fare&＃39;] <&＃61;7.91,&＃39;Fare&＃39;]&＃61;0dataset.loc[(dataset[&＃39;Fare&＃39;] >7.91)&(dataset[&＃39;Fare&＃39;] <&＃61;14.454),&＃39;Fare&＃39;]&＃61;1dataset.loc[(dataset[&＃39;Fare&＃39;] >14.454)&(dataset[&＃39;Fare&＃39;] <&＃61;31),&＃39;Fare&＃39;]&＃61;2dataset.loc[dataset[&＃39;Fare&＃39;] >31,&＃39;Fare&＃39;]&＃61;3dataset[&＃39;Fare&＃39;] &＃61; dataset[&＃39;Fare&＃39;].astype(int)train_df.drop([&＃39;FareBand&＃39;],axis&＃61;1,inplace&＃61;True)
combine&＃61;[train_df,test_df]
train_df


合并特征

将SibSp(兄弟姐妹配偶)和Parch(父母)两组数据合并为一组Family
后借助family判断是否为一个人出行IsAlong

for database in combine:database[&＃39;Family&＃39;]&＃61;database[&＃39;SibSp&＃39;] &＃43; database[&＃39;Parch&＃39;] &＃43; 1# train_df.head()for database in combine:database[&＃39;IsAlong&＃39;]&＃61;1database.loc[database[&＃39;Family&＃39;]>1,&＃39;IsAlong&＃39;]&＃61;0train_df.drop([&＃39;SibSp&＃39;,&＃39;Parch&＃39;,&＃39;Family&＃39;],inplace&＃61;True,axis&＃61;1)
test_df.drop([&＃39;SibSp&＃39;,&＃39;Parch&＃39;,&＃39;Family&＃39;],inplace&＃61;True,axis&＃61;1)train_df


制作模型

#选择变量
X_train &＃61; train_df.drop(&＃39;Survived&＃39;,axis&＃61;1)
Y_train &＃61; train_df[&＃39;Survived&＃39;]


#逻辑回归模型
from sklearn.linear_model import LogisticRegressionlogreg &＃61; LogisticRegression()#实例化 调用
logreg.fit(X_train,Y_train)#fit拟合模型
Y_pred &＃61; logreg.predict(test_df)#利用predict进行预测
acc_log &＃61; round(logreg.score(X_train,Y_train)*100)#round函数--四舍五入 保留两位小数#支持向量机 SVC
from sklearn.svm import SVCsvc &＃61; SVC()
svc.fit(X_train,Y_train)
Y_prd &＃61; svc.predict(test_df)
acc_svc &＃61; round(svc.score(X_train,Y_train)*100,2)#决策树
from sklearn.tree import DecisionTreeClassifierdecision_tree &＃61; DecisionTreeClassifier()
decision_tree.fit(X_train,Y_train)
Y_pred &＃61; decision_tree.predict(test_df)
acc_decision_tree &＃61; round(decision_tree.score(X_train,Y_train)*100,2)#随机森林
from sklearn.ensemble import RandomForestClassifierrandom_forest &＃61; RandomForestClassifier()
random_forest.fit(X_train,Y_train)
Y_pred &＃61; random_forest.predict(test_df)
acc_random_forest &＃61; round(random_forest.score(X_train,Y_train)*100,2)#K阶近邻
from sklearn.neighbors import KNeighborsClassifierknn &＃61; KNeighborsClassifier()
knn.fit(X_train,Y_train)
Y_pred &＃61; knn.predict(test_df)
acc_knn &＃61; round(knn.score(X_train,Y_train)*100,2)#朴素贝叶斯
from sklearn.naive_bayes import GaussianNBgaussian &＃61; GaussianNB()
gaussian.fit(X_train,Y_train)
Y_pred &＃61; gaussian.predict(test_df)
acc_gaussian &＃61; round(gaussian.score(X_train,Y_train)*100,2)


将模型根据评分进行排名

models &＃61; pd.DataFrame({ #采用字典&＃39;Modle&＃39;:[&＃39;Logistic Regression&＃39;,&＃39;Support Vector Machines&＃39;,&＃39;KNN&＃39;,&＃39;Naive Bayes&＃39;,&＃39;Decision Tree&＃39;,&＃39;Random Forest&＃39;],&＃39;Score&＃39;:[acc_log,acc_svc,acc_knn,acc_gaussian,acc_decision_tree,acc_random_forest]
})
models.sort_values(by&＃61;&＃39;Score&＃39;,ascending&＃61;False)


模型预测
有以上模型的排名情况 可选择决策树对数据进行预测

#决策树
from sklearn.tree import DecisionTreeClassifierdecision_tree &＃61; DecisionTreeClassifier()
decision_tree.fit(X_train,Y_train)
Y_pred &＃61; decision_tree.predict(test_df)
acc_decision_tree &＃61; round(decision_tree.score(X_train,Y_train)*100,2)#定义一个新的变量 为预测后的值
submission &＃61; pd.DataFrame({&＃39;PassengerId&＃39;:PassengerId,&＃39;Survived&＃39;:Y_pred
})#将结果输出为文件
submission.to_csv(&＃39;submission.csv&＃39;,index&＃61;False)
submission