Kaggle泰坦尼克号（0.80382，更新中）

我的Kaggle账号：SevenTailCat / SixTailCat

# This Python 3 environment comes with many helpful analytics libraries installed
# It is defined by the kaggle/python docker image: https://github.com/kaggle/docker-python
# For example, here's several helpful packages to load in

import numpy as np # linear algebra
import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)
import math
from sklearn.linear_model import LogisticRegression
from sklearn.svm import SVC, LinearSVC
from sklearn.neighbors import KNeighborsClassifier
from sklearn.naive_bayes import GaussianNB
from sklearn.linear_model import SGDClassifier
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import StratifiedKFold
from sklearn.model_selection import GridSearchCV

# Input data files are available in the "../input/" directory.
# For example, running this (by clicking run or pressing Shift+Enter) will list the files in the input directory

import os
print(os.listdir("../input"))

# Any results you write to the current directory are saved as output.

#合并两个文档
data_train = pd.read_csv('../input/train.csv')
data_test = pd.read_csv('../input/test.csv') #这里读取的是一个读取的Pandas格式
combine = [data_train, data_test] #这是一个列表

#每行缺失值
#data_train.shape[1] - data_train.count(axis=1)
#每列缺失值
data_train.shape[0] - data_train.count()
#这里的结果和我在博客看到的不符，但经过检查，结果无误。猜测：官方更改了数据集

发现内有三种缺失值，分别是：Age，Cabin，Embarked。其中Cabin缺失过大，几乎占总体样本但3/4 。Embarked仅有两个。

#查看预测样本集合内缺失
data_test.shape[0] - data_test.count()

发现内有三种缺失，其中年龄缺失、Cabin缺失和训练集相同，但Fare数据内依然有差别。

data_train.shape

data_train.columns

data_train.dtypes

#此处记录一个非常强悍的查看缺失值的方法
import missingno as msno
msno.matrix(data_train,figsize=(14,6))#figsize制图大小

data_train.head()

Age缺失值的处理方法

方法一:填充中位数

#方法一：填充中位数
age_median = data_train.Age.median()
age_median

方法二：根据每个人对应的称谓中位数填充，更加准确，如miss和Mrs可能年龄不同

#方法二：根据每个人对应的称谓中位数填充，更加准确，如miss和Mrs可能年龄不同
#1.检查称谓
count = 0 #count the number of names without "Mr./Miss./Master./Dr./Mrs."
def check_if_the_name_including_reguler(dataset,count):
    for data in dataset.Name:
        if ("Mr." not in data) & ("Miss." not in data) & ("Master." not in data) & ("Dr." not in data) & ("Mrs." not in data):
            print(data)
            count = count +1
    print("\nthe number of names without 'Mr./Miss./Master./Dr./Mrs.' is " + str(count))
    return count

check_if_the_name_including_reguler(data_train,count)

结果可见，仅20人没有带有相关称谓。根据幸存者手册（https://www.encyclopedia-titanica.org/titanic-survivor/lady-duff-gordon.html） 其中部分是爵、牧师等特殊身份的人。

#统计每种称谓的平均年龄
age_in = []   #Mr./Miss./Master./Dr./Mrs.
for i in range(0,5):
    age_in.insert(0,[0,0])
def count_every_call_mid(data_train,age_in):
    #data_train.ix["Miss." in data_train.Name]
    for data in data_train.index:
        #print(data_train.loc[data,"Name"])
        if "Mr." in data_train.loc[data,"Name"]:
            if not math.isnan(data_train.loc[data,"Age"]) :
                age_in[0][0]+=data_train.loc[data,"Age"]
                age_in[0][1]+=1
        elif "Miss." in data_train.loc[data,"Name"]:
            if not math.isnan(data_train.loc[data,"Age"]) :
                age_in[1][0]+=data_train.loc[data,"Age"]
                age_in[1][1]+=1
        elif "Master." in data_train.loc[data,"Name"]:
            if not math.isnan(data_train.loc[data,"Age"]) :
                age_in[2][0]+=data_train.loc[data,"Age"]
                age_in[2][1]+=1
        elif "Dr." in data_train.loc[data,"Name"]:
            if not math.isnan(data_train.loc[data,"Age"]) :
                age_in[3][0]+=data_train.loc[data,"Age"]
                age_in[3][1]+=1
        elif "Mrs." in data_train.loc[data,"Name"]:
            if not math.isnan(data_train.loc[data,"Age"]) :
                age_in[4][0]+=data_train.loc[data,"Age"]
                age_in[4][1]+=1
    return age_in

count_every_call_mid(data_train,age_in)
age_median = data_train.Age.median()
ave=[0,0,0,0,0]
for i in range(0,5):
    ave[i]=age_in[i][0]/age_in[i][1]
    print(ave[i])#Mr./Miss./Master./Dr./Mrs.

通过统计结果表明，每个称呼的年龄和全称呼结果相差较大。

#将各称呼年龄平均年龄填充，非常用称呼用平均年龄填充
def replace_age_by_mid_of_call(dataset):
    for data in dataset.index:
        #print(data_train.loc[data,"Name"])
        if math.isnan(dataset.loc[data,"Age"]) :
            if "Mr." in dataset.loc[data,"Name"]:
                dataset.loc[data,"Age"]=ave[0]
            elif "Miss." in dataset.loc[data,"Name"]:
                dataset.loc[data,"Age"]=ave[1]
            elif "Master." in dataset.loc[data,"Name"]:
                dataset.loc[data,"Age"]=ave[2]
            elif "Dr." in dataset.loc[data,"Name"]:
                dataset.loc[data,"Age"]=ave[3]
            elif "Mrs." in dataset.loc[data,"Name"]:
                dataset.loc[data,"Age"]=ave[4]
            else:
                dataset.loc[data,"Age"]=age_median
replace_age_by_mid_of_call(data_train)

方法三：将预测样本集和训练样本集在一起处理。

因为在不涉及结果，而对单个特征进行处理时，可以将所有集合看作一个新的样本。通过这个“新样本集合”来预测缺失值。

#此处对于pandas读取的文件和列表有不同的处理格式
#统计整体内除了基本称呼"Mr./Miss./Master./Dr./Mrs."以外的称呼占总体称呼的多少
count = 0
for dataset in combine:
    count = check_if_the_name_including_reguler(dataset,count)
    print("\n")

可见，在总体样本中，除了基本称呼以外并不多。

#统计所有样本中，每种称谓的平均年龄：
age_in = []   #Mr./Miss./Master./Dr./Mrs.
for i in range(0,5):
    age_in.insert(0,[0,0])
for dataset in combine:
    age_in = count_every_call_mid(dataset,age_in)
age_median = data_train.Age.median()
ave=[0,0,0,0,0]
for i in range(0,5):
    ave[i]=age_in[i][0]/age_in[i][1]
    print(ave[i])#Mr./Miss./Master./Dr./Mrs.

由此对比，可见，五个称呼的平均值均有不同程度的修正。再用新的修正值填充空缺Age值。

#用新的平均值填充
def replace_age_by_mid_of_call(dataset):
    for data in dataset.index:
        #print(data_train.loc[data,"Name"])
        if math.isnan(dataset.loc[data,"Age"]) :
            if "Mr." in dataset.loc[data,"Name"]:
                dataset.loc[data,"Age"]=ave[0]
            elif "Miss." in dataset.loc[data,"Name"]:
                dataset.loc[data,"Age"]=ave[1]
            elif "Master." in dataset.loc[data,"Name"]:
                dataset.loc[data,"Age"]=ave[2]
            elif "Dr." in dataset.loc[data,"Name"]:
                dataset.loc[data,"Age"]=ave[3]
            elif "Mrs." in dataset.loc[data,"Name"]:
                dataset.loc[data,"Age"]=ave[4]
            else:
                dataset.loc[data,"Age"]=age_median

for dataset in combine:
    replace_age_by_mid_of_call(dataset)

#检查样本是否被填充
data_test.shape[0] - data_test.count()
#检查样本是否被填充
#data_train.shape[0] - data_train.count()

Cabin处理

#方法一：直接丢弃该特征
#data_train = data_train.drop(['Cabin'],axis=1)
#有时候运行两次会出错，因为同一个特征无法丢弃两次

#方法二：利用
def replaceCabin(data):
    data['Cabin'] = data['Cabin'].fillna('X')
    data['Cabin'] = data['Cabin'].apply(lambda x: str(x)[0])
    data['Cabin'] = data['Cabin'].replace(['A', 'D', 'E', 'T'], 'M')
    data['Cabin'] = data['Cabin'].replace(['B', 'C'], 'H')
    data['Cabin'] = data['Cabin'].replace(['F', 'G'], 'L')
    data['Cabin'] = data['Cabin'].map({'X': 0, 'L': 1, 'M': 2, 'H': 3}).astype(int)

# def replaceCabin(data):
#     data['Cabin'] = data['Cabin'].fillna('X')
#     data['Cabin'] = data['Cabin'].apply(lambda x: str(x)[0])
# #     data['Cabin'] = data['Cabin'].replace(['A', 'D', 'E', 'T'], 'M')
# #     data['Cabin'] = data['Cabin'].replace(['B', 'C'], 'H')
# #     data['Cabin'] = data['Cabin'].replace(['F', 'G'], 'L')
#     data['Cabin'] = data['Cabin'].map({'X': 0, 'A':1,'B':2,'C':3,'D':4,'E':5,'F':6,'G':7,'T':8}).astype(int) 

for dataset in combine:
    replaceCabin(dataset)

Sex处理

# for dataset in combine:
#     print(dataset)
#     if 'male' not in dataset['Sex']:
#          print(dataset['Sex'])
#     dataset['Sex'] = dataset['Sex'].map({'female': 1, 'male': 0}).astype(int)

# data_train.head()
data_train.replace({'Sex':{'female':1,'male':2}},inplace=True)
# #处理预测样本集
data_test.replace({'Sex':{'female':1,'male':2}},inplace=True)

Fare处理

因为Fare仅在预测样本中缺失，因此，不能直接丢弃。

#方法一：选择平均值填充
fare_median = data_test.Fare.median()
data_test['Fare'].fillna(data_test['Fare'].dropna().median(), inplace=True)

#抽取前五个样本，查看是否所有特征均得到合适处理
data_train.head()

#检查预测样本集是否对特征进行了处理
data_test.head()

Tickes处理

# Ticket
df = data_train['Ticket'].value_counts()
df = pd.DataFrame(df)
df = df[df['Ticket'] > 1]
df_ticket = df.index.values        #共享船票的票号
tickets = data_train.Ticket.values    #所有的船票
result = []
for ticket in tickets:
    if ticket in df_ticket:
        ticket = 1
    else:
        ticket = 0                 #遍历所有船票，在共享船票里面的为1，否则为0
    result.append(ticket)
results = pd.DataFrame(result)
results.columns = ['Ticket_Count']
data_train = pd.concat([data_train, results], axis=1)

df = data_test['Ticket'].value_counts()
df = pd.DataFrame(df)
df = df[df['Ticket'] > 1]
df_ticket = df.index.values        #共享船票的票号
tickets = data_test.Ticket.values    #所有的船票
result = []
for ticket in tickets:
    if ticket in df_ticket:
        ticket = 1
    else:
        ticket = 0                 #遍历所有船票，在共享船票里面的为1，否则为0
    result.append(ticket)
results = pd.DataFrame(result)
results.columns = ['Ticket_Count']
data_test = pd.concat([data_test, results], axis=1)

Cabin处理

# Cabin
for dataset in combine:
    dataset.loc[(dataset.Cabin.isnull()), 'Cabin'] = 0
    dataset.loc[(dataset.Cabin.notnull()), 'Cabin'] = 1

Embarked处理

#方法一：因为只有两个，直接丢去
data_train = data_train.dropna(subset=['Embarked'])
# for data in combine:
#     data.loc[data['Embarked']=='S']=1
#     data.loc[data['Embarked']=='C']=2
#     data.loc[data['Embarked']=='Q']=2
data_train = data_train.replace({'Embarked':{'S':1,'C':2,'Q':3}})
data_test = data_test.replace({'Embarked':{'S':1,'C':2,'Q':3}})

data_test.shape[0] - data_test.count()

data_train.shape[0] - data_train.count()

特征

#特征
data_test_X  = data_test[['Pclass', 'Sex', 'Age','SibSp', 'Parch', 'Fare', 'Embarked','Ticket_Count','Cabin']]
X_train = data_train[['Pclass', 'Sex', 'Age','SibSp', 'Parch', 'Fare', 'Embarked','Ticket_Count','Cabin']]
Y_train = data_train['Survived']
X_test = data_test_X

#模型比较
#随机森林
random_forest = RandomForestClassifier(n_estimators=300, max_depth=5, criterion='entropy')
random_forest.fit(X_train, Y_train)
Y_pred = random_forest.predict(X_test)

#模型写出
submission = pd.DataFrame({
        "PassengerId": data_test["PassengerId"],
        "Survived": Y_pred
    })
submission.to_csv('submission.csv', index=False)