kaggle平台学习复习笔记 | 数据划分与模型集成

数据集划分与交叉验证

数据集划分
通常有两种方法：

1. 留出法(Hold-out) 适用于数据量大的情况
2. K折交叉验证(K-fold CV) 适用于数据量一般情况 时间比较长
3. 自助采样(Bootstrap) 较少使用

交叉验证得到的模型更加稳定.

数据一致性分析

理想情况下AUC接近0.5

sklearn中封装的一系列的数据划分的代码

# hold-out
from sklearn.model_selection import train_test_split

# K折交叉验证
from sklearn.model_selection import KFold
from sklearn.model_selection import RepeatedKFold

# K折分布保持交叉验证
from sklearn.model_selection import StratifiedKFold
from sklearn.model_selection import RepeatedStratifiedKFold

# 时间序列划分方法
from sklearn.model_selection import TimeSeriesSplit

# booststrap 采样
from sklearn.utils import resample

构造数据集

X = np.zeros((20, 5))
Y = np.array([1]*5 + [2]*5 + [3]*5 + [4]*5)
print(X, Y)

留出法代码

# 直接按照比例拆分
train_X, val_X, train_y, val_y = train_test_split(X, Y, test_size = 0.2)
print(train_y, val_y)

# 按照比例 & 标签分布划分
train_X, val_X, train_y, val_y = train_test_split(X, Y, test_size = 0.2, stratify=Y)
print(train_y, val_y)

两种划分效果如下：

可以看到，第一种划分不均匀，当添加参数stratify=Y时，
第二种划分方式变的均匀了。

K折交叉验证
不均匀的分布方式 KFold

kf = KFold(n_splits=5)
for train_idx, test_idx, in kf.split(X, Y):
    print(train_idx, test_idx)
    print('Label', Y[test_idx])
    print('')

均匀分布方式 StratifiedKFold

# 5折划分方式
kf = StratifiedKFold(n_splits=5)
for train_idx, test_idx, in kf.split(X, Y):
    print(train_idx, test_idx)
    print('Label', Y[test_idx])
    print('')

均匀重复采样 RepeatedStratifiedKFold

kf = RepeatedStratifiedKFold(n_splits=5, n_repeats=2)
for train_idx, test_idx, in kf.split(X, Y):
    print(train_idx, test_idx)
    print('Label', Y[test_idx])
    print('')

模型集成方法

机器学习的模型集成

1. 模型集成和集成学习
   

2. Vote和Blend
   Vote主要用于分类任务
   Blend主要用于回归任务
   主要方法：
   对结果简单投票或者平均
   加权权重可以修改

3. Stacking
   
   多次学习
   在第一次学习的基础上生成新的特征，
   在将新的特征拼接到训练集上，进行新的学习

深度学习的模型集成

1. 数据扩增
   
   数据集和测试集都可以进行数据扩增

2. snapshot Ensemble
   
   保存多个局部最优解，找到最终最优解

Titanic为例的简单应用

1. Load data

import numpy as np
import pandas as pd

import matplotlib.pyplot as plt
import seaborn as sns
%matplotlib inline

import warnings
warnings.filterwarnings('ignore')
#import train and test CSV files
train_df = pd.read_csv("../input/train.csv")
test_df = pd.read_csv("../input/test.csv")

#take a look at the training data
train_df.describe(include="all")

print(train_df.columns)

train_df.sample(5)

#see a summary of the training dataset
train_df.describe(include = "all")

2. Data Visualization

# 性别和幸存的关系
sns.barplot(x="Sex", y="Survived", data=train_df)

类似的查看其他字段和幸存率的关系

#draw a bar plot of survival by Pclass
sns.barplot(x="Pclass", y="Survived", data=train_df)

#draw a bar plot for SibSp vs. survival
sns.barplot(x="SibSp", y="Survived", data=train_df)

#draw a bar plot for Parch vs. survival
sns.barplot(x="Parch", y="Survived", data=train_df)
plt.show()

根据年龄进行分组，
(-1, 0) 是 Unknown
(0, 5) 是 Baby
(5, 12) 是 Child
(12, 18) 是 Teenager
(18, 24) 是 Student
(24, 35) 是 Young Adult
.。。。。。
查看年龄和幸存的关系

#sort the ages into logical categories
train_df["Age"] = train_df["Age"].fillna(-0.5)
test_df["Age"] = test_df["Age"].fillna(-0.5)
bins = [-1, 0, 5, 12, 18, 24, 35, 60, np.inf]
labels = ['Unknown', 'Baby', 'Child', 'Teenager', 'Student', 'Young Adult', 'Adult', 'Senior']
train_df['AgeGroup'] = pd.cut(train_df["Age"], bins, labels = labels)
test_df['AgeGroup'] = pd.cut(test_df["Age"], bins, labels = labels)
train_df["CabinBool"] = train_df["Cabin"].notnull().astype('int')
test_df["CabinBool"] = test_df["Cabin"].notnull().astype('int')

删除一些没太有用的信息

train = train.drop(['Cabin'], axis = 1)
test = test.drop(['Cabin'], axis = 1)

#we can also drop the Ticket feature since it's unlikely to yield any useful information
train = train.drop(['Ticket'], axis = 1)
test = test.drop(['Ticket'], axis = 1)

处理姓名字段

#replace various titles with more common names
for dataset in combine:
    dataset['Title'] = dataset['Title'].replace(['Lady', 'Capt', 'Col',
    'Don', 'Dr', 'Major', 'Rev', 'Jonkheer', 'Dona'], 'Rare')
    
    dataset['Title'] = dataset['Title'].replace(['Countess', 'Lady', 'Sir'], 'Royal')
    dataset['Title'] = dataset['Title'].replace('Mlle', 'Miss')
    dataset['Title'] = dataset['Title'].replace('Ms', 'Miss')
    dataset['Title'] = dataset['Title'].replace('Mme', 'Mrs')

train[['Title', 'Survived']].groupby(['Title'], as_index=False).mean()

# map each of the title groups to a numerical value
title_mapping = {"Mr": 1, "Miss": 2, "Mrs": 3, "Master": 4, "Royal": 5, "Rare": 6}
for dataset in combine:
    dataset['Title'] = dataset['Title'].map(title_mapping)
    dataset['Title'] = dataset['Title'].fillna(0)

train.head()

登陆口

#now we need to fill in the missing values in the Embarked feature
print("Number of people embarking in Southampton (S):")
southampton = train[train["Embarked"] == "S"].shape[0]
print(southampton)

print("Number of people embarking in Cherbourg (C):")
cherbourg = train[train["Embarked"] == "C"].shape[0]
print(cherbourg)

print("Number of people embarking in Queenstown (Q):")
queenstown = train[train["Embarked"] == "Q"].shape[0]
print(queenstown)

用出现次数最多的S做缺失值填充

#replacing the missing values in the Embarked feature with S
train = train.fillna({"Embarked": "S"})

年龄字段

#create a combined group of both datasets
combine = [train, test]

#extract a title for each Name in the train and test datasets
for dataset in combine:
    dataset['Title'] = dataset.Name.str.extract(' ([A-Za-z]+)\.', expand=False)

pd.crosstab(train['Title'], train['Sex'])

#replace various titles with more common names
for dataset in combine:
    dataset['Title'] = dataset['Title'].replace(['Lady', 'Capt', 'Col',
    'Don', 'Dr', 'Major', 'Rev', 'Jonkheer', 'Dona'], 'Rare')
    
    dataset['Title'] = dataset['Title'].replace(['Countess', 'Lady', 'Sir'], 'Royal')
    dataset['Title'] = dataset['Title'].replace('Mlle', 'Miss')
    dataset['Title'] = dataset['Title'].replace('Ms', 'Miss')
    dataset['Title'] = dataset['Title'].replace('Mme', 'Mrs')

train[['Title', 'Survived']].groupby(['Title'], as_index=False).mean()

处理姓名字段的缺失值

# fill missing age with mode age group for each title
mr_age = train[train["Title"] == 1]["AgeGroup"].mode() #Young Adult
miss_age = train[train["Title"] == 2]["AgeGroup"].mode() #Student
mrs_age = train[train["Title"] == 3]["AgeGroup"].mode() #Adult
master_age = train[train["Title"] == 4]["AgeGroup"].mode() #Baby
royal_age = train[train["Title"] == 5]["AgeGroup"].mode() #Adult
rare_age = train[train["Title"] == 6]["AgeGroup"].mode() #Adult

age_title_mapping = {1: "Young Adult", 2: "Student", 3: "Adult", 4: "Baby", 5: "Adult", 6: "Adult"}

#使用众数来填充缺失值
for x in range(len(train["AgeGroup"])):
    if train["AgeGroup"][x] == "Unknown":
        train["AgeGroup"][x] = age_title_mapping[train["Title"][x]]
        
for x in range(len(test["AgeGroup"])):
    if test["AgeGroup"][x] == "Unknown":
        test["AgeGroup"][x] = age_title_mapping[test["Title"][x]]

性别字段转换成数值

#map each Sex value to a numerical value
sex_mapping = {"male": 0, "female": 1}
train['Sex'] = train['Sex'].map(sex_mapping)
test['Sex'] = test['Sex'].map(sex_mapping)

train.head()

登陆口岸转换为数值

#map each Embarked value to a numerical value
embarked_mapping = {"S": 1, "C": 2, "Q": 3}
train['Embarked'] = train['Embarked'].map(embarked_mapping)
test['Embarked'] = test['Embarked'].map(embarked_mapping)

train.head()

根据票类型的平均值填补缺失值

for x in range(len(test["Fare"])):
    if pd.isnull(test["Fare"][x]):
        pclass = test["Pclass"][x] #Pclass = 3
        test["Fare"][x] = round(train[train["Pclass"] == pclass]["Fare"].mean(), 4)
        
# map Fare values into groups of numerical values
train['FareBand'] = pd.qcut(train['Fare'], 4, labels = [1, 2, 3, 4])
test['FareBand'] = pd.qcut(test['Fare'], 4, labels = [1, 2, 3, 4])

train = train.drop(['Fare'], axis = 1)
test = test.drop(['Fare'], axis = 1)

查看数据集结果

#check train data
train.head()

可以看到数据均被转换为合理的数值格式，供模型训练使用

3. Choosing the Best Model

from sklearn.model_selection import train_test_split

predictors = train.drop(['Survived', 'PassengerId'], axis=1)
target = train["Survived"]
x_train, x_val, y_train, y_val = train_test_split(predictors, target, test_size = 0.22, random_state = 0)

高斯朴素贝叶斯

# Gaussian Naive Bayes
from sklearn.naive_bayes import GaussianNB
from sklearn.metrics import accuracy_score

gaussian = GaussianNB()
gaussian.fit(x_train, y_train)
y_pred = gaussian.predict(x_val)
acc_gaussian = round(accuracy_score(y_pred, y_val) * 100, 2)
print(acc_gaussian)

逻辑回归

# Logistic Regression
from sklearn.linear_model import LogisticRegression

logreg = LogisticRegression()
logreg.fit(x_train, y_train)
y_pred = logreg.predict(x_val)
acc_logreg = round(accuracy_score(y_pred, y_val) * 100, 2)
print(acc_logreg)

svm的分类模型

# Support Vector Machines
from sklearn.svm import SVC

svc = SVC()
svc.fit(x_train, y_train)
y_pred = svc.predict(x_val)
acc_svc = round(accuracy_score(y_pred, y_val) * 100, 2)
print(acc_svc)

线性SVC模型

# Linear SVC
from sklearn.svm import LinearSVC

linear_svc = LinearSVC()
linear_svc.fit(x_train, y_train)
y_pred = linear_svc.predict(x_val)
acc_linear_svc = round(accuracy_score(y_pred, y_val) * 100, 2)
print(acc_linear_svc)

线性模型

# Perceptron
from sklearn.linear_model import Perceptron

perceptron = Perceptron()
perceptron.fit(x_train, y_train)
y_pred = perceptron.predict(x_val)
acc_perceptron = round(accuracy_score(y_pred, y_val) * 100, 2)
print(acc_perceptron)

决策树模型

#Decision Tree
from sklearn.tree import DecisionTreeClassifier

decisiontree = DecisionTreeClassifier()
decisiontree.fit(x_train, y_train)
y_pred = decisiontree.predict(x_val)
acc_decisiontree = round(accuracy_score(y_pred, y_val) * 100, 2)
print(acc_decisiontree)

随机森林

# Random Forest
from sklearn.ensemble import RandomForestClassifier

randomforest = RandomForestClassifier()
randomforest.fit(x_train, y_train)
y_pred = randomforest.predict(x_val)
acc_randomforest = round(accuracy_score(y_pred, y_val) * 100, 2)
print(acc_randomforest)

KNN近邻模型

# KNN or k-Nearest Neighbors
from sklearn.neighbors import KNeighborsClassifier

knn = KNeighborsClassifier()
knn.fit(x_train, y_train)
y_pred = knn.predict(x_val)
acc_knn = round(accuracy_score(y_pred, y_val) * 100, 2)
print(acc_knn)

随机梯度下降

# Stochastic Gradient Descent
from sklearn.linear_model import SGDClassifier

sgd = SGDClassifier()
sgd.fit(x_train, y_train)
y_pred = sgd.predict(x_val)
acc_sgd = round(accuracy_score(y_pred, y_val) * 100, 2)
print(acc_sgd)

梯度提升决策树

# Gradient Boosting Classifier
from sklearn.ensemble import GradientBoostingClassifier

gbk = GradientBoostingClassifier()
gbk.fit(x_train, y_train)
y_pred = gbk.predict(x_val)
acc_gbk = round(accuracy_score(y_pred, y_val) * 100, 2)
print(acc_gbk)

模型评分

models = pd.DataFrame({
    'Model': ['Support Vector Machines', 'KNN', 'Logistic Regression', 
              'Random Forest', 'Naive Bayes', 'Perceptron', 'Linear SVC', 
              'Decision Tree', 'Stochastic Gradient Descent', 'Gradient Boosting Classifier'],
    'Score': [acc_svc, acc_knn, acc_logreg, 
              acc_randomforest, acc_gaussian, acc_perceptron,acc_linear_svc, acc_decisiontree,
              acc_sgd, acc_gbk]})
              
models.sort_values(by='Score', ascending=False)

算法评分结果如下：

kaggle比赛相关tips