kaggle 机器学习 随机森林

kaggle项目：使用随机森林模型进行泰坦尼克号生存预测

Posted by lastlorder on April 3, 2019

泰坦尼克号是一个入门的机器学习项目，通过这个项目了解机器学习的基本方法和数据分析的方法 #首先进行数据的分析和清洗

import re

import numpy as np

import pandas as pd

import matplotlib.pyplot as plt

import seaborn as sns

import warnings

warnings.filterwarnings("ignore")

# 导入数据

train_data = pd.read_csv(r'C:\Users\Administrator\Desktop\taitanic\train.csv')

test_data = pd.read_csv(r'C:\Users\Administrator\Desktop\taitanic\test.csv')

# 总体数据信息

train_data.info()

test_data.info()

sum_data = train_data.append(test_data)

psg_out_id = test_data['PassengerId']
# 空余上船位置定义

#train_data['Cabin'] = train_data.Cabin.fillna('U0') 
# 开始数据分析
# 总体生存比例
train_data['Survived'].value_counts().plot.pie(autopct='%0.9f%%') 
# 性别和生存关系
print(train_data.groupby(['Sex','Survived'])['Survived'].count())
# 船舱，性别与生存的关系
print(train_data.groupby(['Sex','Pclass','Survived'])['Survived'].count())
# 作图 性别和生存
train_data[['Sex','Survived']].groupby(['Sex']).mean().plot.bar()
# 船舱与生存的关系
train_data[['Pclass','Survived']].groupby(['Pclass']).mean().plot.bar()
# 船舱，性别与生存的关系作图
train_data[['Sex','Pclass','Survived']].groupby(['Pclass','Sex']).mean().plot.bar()

# 其次看一下仓位因素

fig = plt.figure()

Us =train_data.Pclass[train_data.Survived == 0].value_counts()

S = train_data.Pclass[train_data.Survived == 1].value_counts()

df = pd.DataFrame({'Survived':S, 'Unsurvived':Us})

df.plot(kind='bar', stacked=True)

plt.title("The rescue of each passenger level ")

plt.xlabel("Level") 

plt.ylabel("Number of people") 

plt.xticks(rotation = 0)

# 仓位与年龄的关系

Us =train_data.Age[train_data.Survived == 0].value_counts()

S = train_data.Age[train_data.Survived == 1].value_counts()

df = pd.DataFrame({'Survived':S, 'Unsurvived':Us})

df.plot(kind='bar', stacked=True)

plt.title("Age to Survrive ")

plt.xlabel("Age") 

plt.ylabel("Number of people") 

plt.xticks(rotation = 0)

# 港口与生存的关系

sns.countplot('Embarked',hue='Survived',data=train_data)

plt.title('Embarked and Survived')

#回归森林预测年龄
from sklearn.ensemble import RandomForestRegressor

# 通过其他属性进行分析

age_df = sum_data[['Age','Title','Parch', 'SibSp','Pclass']]

age_df_notnull = age_df.loc[(sum_data['Age'].notnull())]

age_df_isnull = age_df.loc[(sum_data['Age'].isnull())]

X = age_df_notnull.values[:,1:]

Y = age_df_notnull.values[:,0]

# 算法选择随机森林，当然也可以是其他的

RFR = RandomForestRegressor(n_estimators=1000, n_jobs=-1)

RFR.fit(X,Y)

predictAges = RFR.predict(age_df_isnull.values[:,1:])

sum_data.loc[sum_data['Age'].isnull(), ['Age']]= predictAges

# 年龄分组

主要分成四组

bins = [0,18, 30, 60, 150]

sum_data['Age_group'] =sum_data['Age']

sum_data.loc[sum_data['Age_group'] >60,'Age_group'] = -4

sum_data.loc[sum_data['Age_group'] >30,'Age_group'] = -3

sum_data.loc[sum_data['Age_group'] >18,'Age_group'] = -2

sum_data.loc[sum_data['Age_group'] >0,'Age_group'] = 1

sum_data.loc[sum_data['Age_group'] ==-2,'Age_group'] = 2

sum_data.loc[sum_data['Age_group'] ==-3,'Age_group'] = 3

sum_data.loc[sum_data['Age_group'] ==-4,'Age_group'] = 4

pd.concat([sum_data,sum_data['Age_group']])

# 亲友人数和生存的关系

n,ax=plt.subplots(1,2,figsize=(18,8))

train_data[['Parch','Survived']].groupby(['Parch']).mean().plot.bar(ax=ax[0])

ax[0].set_title('Parch and Survived')

train_data[['SibSp','Survived']].groupby(['SibSp']).mean().plot.bar(ax=ax[1])

ax[1].set_title('SibSp and Survived')

# 称呼与生存的关系
sum_data = train_data.append(test_data)

sum_data['Title'] = sum_data['Name'].map(lambda x: re.compile(", (.*?)\.").findall(x)[0])

title_Dict = {}
# 称呼分类
title_Dict.update(dict.fromkeys(['Capt', 'Col', 'Major', 'Dr', 'Rev'], 'Officer'))

title_Dict.update(dict.fromkeys(['Don', 'the Countess', 'Dona', 'Lady'], 'Royalty'))

title_Dict.update(dict.fromkeys(['Mme', 'Ms', 'Mrs','Mlle', 'Miss'], 'Mrs'))

title_Dict.update(dict.fromkeys(['Master','Mr', 'Sir'], 'Mr'))

title_Dict.update(dict.fromkeys(['Jonkheer'], 'Master'))

sum_data['Title'] = sum_data['Title'].map(title_Dict)

sum_data['Title'] = pd.factorize(sum_data['Title'])[0]
# 加入tittle
title_dummies_df = pd.get_dummies(sum_data['Title'], prefix=sum_data[['Title']].columns[0])

sum_data = pd.concat([sum_data, title_dummies_df], axis=1)'''
# 对仓位进行简单处理 

sum_data['Cabin'].loc[(sum_data['Cabin'].notnull())] = 1

sum_data['Cabin'].loc[(sum_data['Cabin'].isnull())] = 0

train_data =sum_data[:891]

train_data[['Title','Survived']].groupby(['Title']).mean().plot.bar()

train_data[['Cabin','Survived']].groupby(['Cabin']).mean().plot.bar()

# 作图

plt.show()

搞事的分割线

通过之前的作图，对这个数据有了一定的了解了，可以开始干活了

import pandas as pd 
import  numpy as np 
import re
import matplotlib.pyplot as plt
import seaborn as sns
import datetime
train_data=pd.read_csv(r'C:\Users\Administrator\Desktop\taitanic\train.csv')      #读取文件titanic_train数据集
test_data = pd.read_csv(r'C:\Users\Administrator\Desktop\taitanic\test.csv')
sum_data = train_data.append(test_data)
psg_out_id = test_data['PassengerId']

##名字开头
sum_data['Title'] = sum_data['Name'].map(lambda x: re.compile(", (.*?)\.").findall(x)[0])
title_Dict = {}
title_Dict.update(dict.fromkeys(['Capt', 'Col', 'Major', 'Dr', 'Rev'], 'Officer'))
title_Dict.update(dict.fromkeys(['Don', 'the Countess', 'Dona', 'Lady'], 'Royalty'))
title_Dict.update(dict.fromkeys(['Mme', 'Ms', 'Mrs','Mlle', 'Miss'], 'Mrs'))
title_Dict.update(dict.fromkeys(['Master','Mr', 'Sir'], 'Mr'))
title_Dict.update(dict.fromkeys(['Jonkheer'], 'Master'))
sum_data['Title'] = sum_data['Title'].map(title_Dict)
sum_data['Title'] = pd.factorize(sum_data['Title'])[0]

#亲友
sum_data['FAF']= sum_data['SibSp'] + sum_data['Parch']
pd.concat([sum_data,sum_data['FAF']])
#性别替换
sum_data.loc[sum_data['Sex']=='male','Sex'] = 0
sum_data.loc[sum_data['Sex']=='female','Sex'] = 1
#上船位置
sum_data['Embarked'] = sum_data['Embarked'].fillna('S')
sum_data.loc[sum_data['Embarked'] == 'S','Embarked'] =0
sum_data.loc[sum_data['Embarked'] == 'C','Embarked'] = 1
sum_data.loc[sum_data['Embarked'] == 'Q','Embarked'] = 1
#船舱
sum_data['Cabin'].loc[(sum_data['Cabin'].notnull())] = 1
sum_data['Cabin'].loc[(sum_data['Cabin'].isnull())] = 0
#票
sum_data['Fare'] = sum_data[['Fare']].fillna(sum_data.groupby('Pclass').transform(np.mean))

#等级

sum_data.loc[sum_data['Pclass'] == 1,'Pclass'] =1
sum_data.loc[sum_data['Pclass'] == 2,'Pclass'] = 0
sum_data.loc[sum_data['Pclass'] == 3,'Pclass'] = -1

#回归森林预测年龄
from sklearn.ensemble import RandomForestRegressor
#choose training data to predict age
age_df = sum_data[['Age','Title','Parch', 'SibSp','Pclass']]
age_df_notnull = age_df.loc[(sum_data['Age'].notnull())]
age_df_isnull = age_df.loc[(sum_data['Age'].isnull())]
X = age_df_notnull.values[:,1:]
Y = age_df_notnull.values[:,0]
# use RandomForestRegression to train data
RFR = RandomForestRegressor(n_estimators=1000, n_jobs=-1)
RFR.fit(X,Y)
predictAges = RFR.predict(age_df_isnull.values[:,1:])
sum_data.loc[sum_data['Age'].isnull(), ['Age']]= predictAges

#年龄分组
bins = [0, 12, 18, 65, 100]
sum_data['Age_group'] =sum_data['Age']
sum_data.loc[sum_data['Age_group']  > 65,'Age_group'] = -4
sum_data.loc[sum_data['Age_group'] >30,'Age_group'] = -3
sum_data.loc[sum_data['Age_group'] >15,'Age_group'] = -2
sum_data.loc[sum_data['Age_group'] >0,'Age_group'] = 1
sum_data.loc[sum_data['Age_group'] ==-2,'Age_group'] = 0
sum_data.loc[sum_data['Age_group'] ==-3,'Age_group'] = 0
sum_data.loc[sum_data['Age_group'] ==-4,'Age_group'] = -1
pd.concat([sum_data,sum_data['Age_group']])
#年龄归一化
from sklearn import preprocessing
assert np.size(sum_data['Age']) == 1309
# StandardScaler will subtract the mean from each value then scale to the unit variance
scaler = preprocessing.StandardScaler()
sum_data['Age_scaled'] = scaler.fit_transform(sum_data['Age'].values.reshape(-1,1))
assert np.size(sum_data['Fare']) == 1309
scaler = preprocessing.StandardScaler()
sum_data['Fare_scaled'] = scaler.fit_transform(sum_data['Fare'].values.reshape(-1,1))


#随机森林决策树
from sklearn import model_selection
from sklearn.model_selection import KFold
from sklearn.ensemble import RandomForestClassifier
#两组数据分开
traindata =sum_data[:891]
testdata = sum_data[891:]
#这里后面的samples不要少写一个“s”
#n_estimators是决策树的个数，min_samples_split是根据属性划分节点时，每个划分最少的样本数。min_samples_leaf:叶子节点最少的样本数。


#交叉验证
from sklearn import model_selection
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import cross_val_score
#predictors = [ 'FAF','SibSp',  'Age','Embarked','Fare','Cabin', 'Pclass','Sex', 'Title']     0.811
#predictors = [ 'FAF', 'Age','Embarked','Fare','Cabin', 'Pclass','Sex', 'Title']                  0.818
#predictors = [ 'FAF','SibSp',  'Age','Embarked','Fare','Cabin','Sex', 'Title']                   0.816
#predictors = ['Sex', 'Title', 'FAF','SibSp',  'Age','Embarked','Fare','Cabin', 'Pclass']     0.817
#predictors = ['Sex', 'Title', 'FAF','SibSp','Embarked','Fare','Cabin', 'Pclass']                0.804
#predictors = ['Sex', 'Title', 'FAF',  'Age','Embarked','Fare','Cabin', 'Pclass']                 0.817
#predictors = ['Sex', 'Title', 'FAF', 'Age','Embarked','Fare','Cabin']                                0.811
#predictors = ['Sex', 'Age', 'Title', 'FAF','Embarked','Fare','Cabin', 'Pclass']                  0.816
#predictors = ['Sex', 'Title', 'FAF','Cabin',  'Age','Embarked','Fare', 'Pclass']                 0.814
#predictors = ['Sex', 'Pclass', 'Title', 'FAF',  'Age','Embarked','Fare','Cabin']                 0.814
#predictors = [ 'Pclass','Sex',  'Age','Cabin','Title']                                                   0.795
#predictors =['Sex','Title','Age','FAF', 'Pclass','Embarked','Cabin','Fare']                      0.819
#predictors =['Sex','Title','Age','FAF', 'Pclass','Embarked','Cabin','Fare','Parch']          0.821     
predictors =['Sex','Title','Age','FAF', 'Pclass','Embarked','Cabin','Fare','Parch']            #0.825 16 1 |0.824 16 2| 0.826 6 2|0.828 |5 2 |4 2 |5 3|0.830 10 1|0.83170  9 1|0.83168 8 2|0.8328 9 2 

alg = RandomForestClassifier(random_state=1,n_estimators = 100,min_samples_split=6,min_samples_leaf=2)
scores = cross_val_score(alg,traindata[predictors],traindata["Survived"],cv=10)
#生成成绩和平均分
print(scores)
print(scores.mean())

#predictors = ['Sex','Title','Age','Age_group','FAF','Embarked','Cabin','Fare']
#predictors = ['Sex','Title','Age_scaled','Age_group','FAF','Embarked','Cabin','Fare_scaled']
#predictors = ['Pclass','Age_scaled','Age_group','SibSp', 'Parch','Fare','Embarked','Title','Cabin','FAF','Sex']
predictors = ['Pclass','Age_scaled','Age_group','SibSp', 'Parch','Fare','Embarked','Title','Cabin','FAF','Sex']#选定特征
UseFlag = traindata['Survived'].values
UseFeature = traindata[predictors].values
rf = RandomForestClassifier(random_state=1,n_estimators=100,min_samples_split=9,min_samples_leaf=2)
rf.fit(UseFeature,UseFlag)#进行模型的训练  

#开始预测
temp = rf.predict(UseFeature)
TestFeature = testdata[predictors].values

temp = rf.predict(TestFeature)
testdata['Survived'] = temp
outdata = testdata[['PassengerId','Survived']]#提取出需要的列
outdata.to_csv(r"C:\Users\Administrator\Desktop\taitanic\RF5.csv",index=False,header=True)#保存数据集

结果在0.78904和0.79904浮动

RF1:年龄和票取众数 0.74

RF2：年龄采用随机森林补充，票取众数，年龄归一化 0.745

RF3: 年龄采用随机森林补充，票取众数，年龄 票 归一化 0.755

RF4: 年龄采用随机森林补充，票取众数，年龄 票 归一化 船舱二分类 加入亲友合计,姓名关键提取，0.77

RF5: 年龄采用随机森林补充，票取众数，年龄 票 归一化 船舱二分类 加入亲友合计，姓名关键提取，加入年龄分段 0.799

随着加入的参数增加，结果也越来越好，但是Pclass作为一个相对关键的因素，在加入后的效果并不如预期，删除后反而准确率更高，由于在之前组中的Pclass的权重过高，因此应适当降低其权重，或者提高其他特征的占比。