Titanic Kaggle 竞赛系列

Titanic: Machine Learning from Disaster

Kaggle 竞赛中的练习题，题目链接在这里。

解题报告源码链接在这里。

Accuracy: 0.77990

Description

The sinking of the RMS Titanic is one of the most infamous shipwrecks in history. On April 15, 1912, during her maiden voyage, the Titanic sank after colliding with an iceberg, killing 1502 out of 2224 passengers and crew. This sensational tragedy shocked the international community and led to better safety regulations for ships.

One of the reasons that the shipwreck led to such loss of life was that there were not enough lifeboats for the passengers and crew. Although there was some element of luck involved in surviving the sinking, some groups of people were more likely to survive than others, such as women, children, and the upper-class.

In this challenge, we ask you to complete the analysis of what sorts of people were likely to survive. In particular, we ask you to apply the tools of machine learning to predict which passengers survived the tragedy.

Tutorials

This Kaggle “Getting Started” Competition provides an ideal starting place for people who may not have a lot of experience in data science and machine learning. The data is highly structured, and we provide tutorials of increasing complexity for using Excel, Python, pandas in Python, and a Random Forest in Python (see links in the sidebar). We also have links to tutorials using R instead. Please use the forums freely and as much as you like. There is no such thing as a stupid question; we guarantee someone else will be wondering the same thing!

More tutorials in the forum.

Answer

0-1 分类：
根据乘客类型，姓名，性别，年龄，兄弟个数，父子个数，船票，票价，船舱，港口

判定乘客是船难中存活下来

数据分析与处理：

需要对 港口，年龄，船票 项做缺失项补全

对 性别，港口 项做字典替换

除去无用信息

利用 Pandas 工具包处理数据

分类：

可通过组合加入新的特征，即人工特征

可以对特征做标准化处理

多种分类方法同时判定，然后加权获得判定结果

总结：

增加特征有利于分类

随机森林在分类上确实很好用

组合分类方法，有点作用

Source

# Python version: 3.4
# Package scikit-learn, pandas
import pandas as pd
import numpy as np
import csv as csv
from sklearn.ensemble import RandomForestClassifier, AdaBoostClassifier
from sklearn.svm import SVC
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression
from sklearn.cross_validation import train_test_split
from sklearn.cross_validation import cross_val_score
from sklearn import cross_validation  #导入交叉检验

# 数据缺失项补齐的设定
#     df.Embarked[df.Embarked.isnull()] = df.Embarked.dropna().mode().values
pd.set_option('chained_assignment',None)

#######################################################################################
# 数据集读取，return 数据集，编号，港口字典
def csvFileRead(filename, Ports_dict):

# pandas 的数据读取
df = pd.read_csv(filename, header=0)

# Sex 中 'female':0, 'male':1 替换，并生成新的一列 Gender
df['Gender'] = df['Sex'].map({'female':0, 'male':1}).astype(int)

# 对 Embarked 项做补缺失项操作
if len(df.Embarked[df.Embarked.isnull()]) > 0:
df.Embarked[df.Embarked.isnull()] = df.Embarked.dropna().mode().values

# 建立 Embarked 项字典，并映射为 ID, 'S','Q','C'
if not Ports_dict:
Ports = list(enumerate(np.unique(df['Embarked'])))
Ports_dict = {name : i-1 for i, name in Ports}
df.Embarked = df.Embarked.map(lambda x: Ports_dict[x]).astype(int)

# Age 缺失项补齐为平均年龄
median_age = df['Age'].dropna().median()
if len(df.Age[df.Age.isnull()]) > 0:
df.loc[(df.Age.isnull()), 'Age'] = median_age

# Fare 缺失项补齐为各类船票的均值
if len(df.Fare[df.Fare.isnull()]) > 0:
median_fare = np.zeros(3)
for f in range(0,3):
median_fare[f] = df[df.Pclass == f+1]['Fare'].dropna().median()
for f in range(0,3):
df.loc[(df.Fare.isnull()) & (df.Pclass == f+1), 'Fare'] = median_fare[f]

# 提取 PassengerId 项
ids = df['PassengerId'].values

# 舍去无用特征
df = df.drop(['Name', 'Sex', 'Ticket', 'Cabin', 'PassengerId'], axis=1)

return df, ids, Ports_dict
#######################################################################################
# 结果存储
def csvSave(filename, ids, predicted):
with open('result.csv', 'w') as mycsv:
mywriter = csv.writer(mycsv)
mywriter.writerow(["PassengerId","Survived"])
mywriter.writerows(zip(ids, predicted))

#######################################################################################
# 测试对比
def resultCompare(predicted):
file1 = 'gendermodel.csv'
file2 = 'gendermodel.csv'
df1 = pd.read_csv(file1, header=0)
df2 = pd.read_csv(file2, header=0)
result1 = df1.Survived.values
result2 = df2.Survived.values
accuracy1 = 1.0 - sum(abs(result1-predicted))/len(predicted)
accuracy2 = 1.0 - sum(abs(result2-predicted))/len(predicted)
print('[1]: ', accuracy1, '\n[2]: ', accuracy2)

#######################################################################################
# 分类
def classificationResult(train_df, test_df):
train_X = train_df.values[:,1:]
train_Y = train_df.values[:,0]
test_X = test_df.values

print('添加人工特征...')
temp1 = train_X[:,2]*train_X[:,3]
temp2 = train_X[:,4]/train_X[:,1]
temp3 = train_X[:,5]*train_X[:,4]
temp4 = train_X[:,0]*train_X[:,4]
train_X = np.c_[train_X, temp1, temp2, temp3, temp4]

temp1 = test_X[:,2]*test_X[:,3]
temp2 = test_X[:,4]/test_X[:,1]
temp3 = test_X[:,5]*test_X[:,4]
temp4 = test_X[:,0]*test_X[:,4]
test_X = np.c_[test_X, temp1, temp2, temp3, temp4]

print('特征标准化...')
scaler = StandardScaler().fit(train_X)
train_X = scaler.transform(train_X)
test_X = scaler.transform(test_X)

print('训练中...')
classifier = RandomForestClassifier(n_estimators=100)
clf1 = LogisticRegression(C=10000)
clf2 = AdaBoostClassifier()

#交叉检验
#scores=cross_validation.cross_val_score(classifier,train_X,train_Y,cv=5)
#print(scores,scores.mean())

classifier.fit(train_X, train_Y)
clf1.fit(train_X, train_Y)
clf2.fit(train_X, train_Y)

print('预测中...')
predicted = 0.6*classifier.predict(test_X).astype(int)
predicted += 0.2*clf1.predict(test_X).astype(int)
predicted += 0.2*clf2.predict(test_X).astype(int)

return (predicted >= 0.75).astype(int)

#######################################################################################
# main
train_df, train_ids, Ports_dict = csvFileRead('train.csv', {})
test_df, ids, Ports_dict = csvFileRead('test.csv', Ports_dict)
predicted = classificationResult(train_df, test_df)
csvSave('result.csv', ids, predicted)

print('完成.')

resultCompare(predicted)

Future

该竞赛的数据集相对很少，初步清理过后的数据集特征量也很少，要提高分类的准确率，对各特征做一定的分析，我想会有很大的帮助，这里有相关的特征重要性分析。