之前有写过一篇关于Titanic比赛的,这几天上kaggle-Titanic的kernels在MostVost找了一篇排第一的kernels来看,参考链接,这个Kernels在模型方面做得特别好,所以,另写一篇作为总结。
流程
1.观察数据,我们要对数据有所了解,可以参考我的
2.特征工程以及数据清洗
3.跑模型
代码分析
首先,导入我们需要用到的库
import pandas as pd
import numpy as np
from sklearn.cross_validation import KFold
import re
import plotly.graph_objs as go
import plotly.offline as py
from sklearn.ensemble import (RandomForestClassifier, AdaBoostClassifier,
GradientBoostingClassifier, ExtraTreesClassifier)
from sklearn.svm import SVC
import xgboost as xgb
import warnings
warnings.filterwarnings('ignore') # 忽略warning
pd.set_option('display.max_columns', None) # 输出结果显示全部列
然后,导入数据
train = pd.read_csv('train.csv')
test = pd.read_csv('test.csv')
PassengerId = test['PassengerId']
full_data = [train, test]
接下来,我们可以查看我们的数据
# 查看train集的数据
print(train.describe()) # 查看描述性统计,只能看数值型数据。
print(train.info()) # 查看数据的信息
# print(train.head()) # 查看train的前n行数据,默认为前5行
从图上我们可以看到,其中有5列不是数值型的,我们需要对其进行转换成数值,而且Age、Cabin这两列是有缺失值的,我们要对其进行填充或者丢弃。
特征工程以及数据清洗
添加一些新的特征
# 添加新的特征,名字的长度
train['Name_length'] = train['Name'].apply(len)
test['Name_length'] = test['Name'].apply(len)
# 乘客在船上是否有船舱
train['Has_Cabin'] = train["Cabin"].apply(lambda x: 0 if type(x) == float else 1)
test['Has_Cabin'] = test["Cabin"].apply(lambda x: 0 if type(x) == float else 1)
# 结合SibSp和Parch创建新的特性FamilySize
for dataset in full_data:
dataset['FamilySize'] = dataset['SibSp'] + dataset['Parch'] + 1
基于特征FamilySize创建新的特征IsAlone,因为一个人的话,顾虑没有那么多,只需要管好自己,生存的几率会大点,其中又分‘male’和‘female’,因为我记得电影中是有这样的一句台词“让女人和小孩先走”,所以,我们有理由相信,女性的生存率会比男性的要高。
for dataset in full_data:
dataset['IsAlone'] = 0
dataset.loc[(dataset['FamilySize'] == 1) & (dataset['Sex'] == 'male'), 'IsAlone'] = 1
dataset.loc[(dataset['FamilySize'] == 1) & (dataset['Sex'] == 'female'), 'IsAlone'] = 2
通过name,添加特征Title
# 定义从乘客名中提取新的特征[Title]的函数
def get_title(name):
title_search = re.search(' ([A-Za-z]+)\.', name)
# 如果title存在,提取并返回它。
if title_search:
return title_search.group(1)
return ""
# 创建一个新的特征[Title]
for dataset in full_data:
dataset['Title'] = dataset['Name'].apply(get_title)
# 将所有不常见的Title分组为一个“Rare”组
for dataset in full_data:
dataset['Title'] = dataset['Title'].replace(
['Lady', 'Countess', 'Capt', 'Col', 'Don', 'Dr', 'Major', 'Rev', 'Sir', 'Jonkheer', 'Dona'], 'Rare')
dataset['Title'] = dataset['Title'].replace('Mlle', 'Miss')
dataset['Title'] = dataset['Title'].replace('Ms', 'Miss')
dataset['Title'] = dataset['Title'].replace('Mme', 'Mrs')
缺失值填充
- Embarked只缺了两个,所以通过统计三个登船地点,选出了登船人数最多的登船地点(s)来填充。
- Test集的Fare只有一个缺失,所以用了中位数来填充
- Age缺失的比较多,所以在[age_avg - age_std, age_avg + age_std]这个范围取值来填充(其中age_avg是Age的平均值,age_std是Age的标准差)
# 通过统计三个登船地点人数最多的填充缺失值
for dataset in full_data:
dataset['Embarked'] = dataset['Embarked'].fillna('S')
# 缺失值填充,Test集的Fare有一个缺失,按中位数来填充,以及创建一个新的特征[CategoricalFare]
for dataset in full_data:
dataset['Fare'] = dataset['Fare'].fillna(train['Fare'].median())
train['CategoricalFare'] = pd.qcut(train['Fare'], 4)
# 缺失值填充,以及创建新的特征[CategoricalAge]
for dataset in full_data:
age_avg = dataset['Age'].mean()
age_std = dataset['Age'].std()
age_null_count = dataset['Age'].isnull().sum()
age_null_random_list = np.random.randint(age_avg - age_std, age_avg + age_std, size=age_null_count)
dataset['Age'][np.isnan(dataset['Age'])] = age_null_random_list
dataset['Age'] = dataset['Age'].astype(int)
通过Age,创建新的特征,一会用来给Age分组
train['CategoricalAge'] = pd.cut(train['Age'], 5)
print(train['CategoricalAge'])
从图片可以看出,年龄分为了5个范围,所以一会把年龄分为5组(0-4)。
分组以及转换数值
Sex:把性别转为0和1.
Embarked:把登船地点转为0、1、2.
Fare:把费用分为4组
Age:把年龄分为5组
for dataset in full_data:
dataset['Sex'] = dataset['Sex'].map({'female': 0, 'male': 1}).astype(int)
title_mapping = {"Mr": 1, "Miss": 2, "Mrs": 3, "Master": 4, "Rare": 5}
dataset['Title'] = dataset['Title'].map(title_mapping)
dataset['Title'] = dataset['Title'].fillna(0)
dataset['Embarked'] = dataset['Embarked'].map({'S': 0, 'C': 1, 'Q': 2}).astype(int)
dataset.loc[dataset['Fare'] <= 7.91, 'Fare'] = 0
dataset.loc[(dataset['Fare'] > 7.91) & (dataset['Fare'] <= 14.454), 'Fare'] = 1
dataset.loc[(dataset['Fare'] > 14.454) & (dataset['Fare'] <= 31), 'Fare'] = 2
dataset.loc[dataset['Fare'] > 31, 'Fare'] = 3
dataset['Fare'] = dataset['Fare'].astype(int)
dataset.loc[dataset['Age'] <= 16, 'Age'] = 0
dataset.loc[(dataset['Age'] > 16) & (dataset['Age'] <= 32), 'Age'] = 1
dataset.loc[(dataset['Age'] > 32) & (dataset['Age'] <= 48), 'Age'] = 2
dataset.loc[(dataset['Age'] > 48) & (dataset['Age'] <= 64), 'Age'] = 3
dataset.loc[dataset['Age'] > 64, 'Age'] = 4
特征选择,丢弃一些不必要的特征
drop_elements = ['PassengerId', 'Name', 'Ticket', 'Cabin', 'SibSp']
train = train.drop(drop_elements, axis=1)
train = train.drop(['CategoricalAge', 'CategoricalFare'], axis=1)
test = test.drop(drop_elements, axis=1)
# print(train.head())
print(train.describe())
# print(train.head())
跑模型
这部分是这个kernels的重点,用的是Stacking。Stacking使用第一级分类器的预测作为对第二级模型的训练输入。我们使用了(RandomForestClassifier, AdaBoostClassifier,GradientBoostingClassifier, ExtraTreesClassifier,Support Vector Classifier)这5个分类器的预测作为下一个分类器(xgboost)的特征。
在下面的代码中,我们编写了一个类SklearnHelper,它允许扩展所有Sklearn分类器所共有的内置方法(如train、predict和fit)。这消除了冗余,因为如果我们想调用5个不同的分类器,就不需要编写相同的方法5次。
# 一些有用的参数,下面会用到
ntrain = train.shape[0]
ntest = test.shape[0]
SEED = 0
NFOLDS = 5
kf = KFold(ntrain, n_folds=NFOLDS, random_state=SEED)
class SklearnHelper(object):
def __init__(self, clf, seed=0, params=None):
params['random_state'] = seed
self.clf = clf(**params)
def train(self, x_train, y_train):
self.clf.fit(x_train, y_train)
def predict(self, x):
return self.clf.predict(x)
def fit(self, x, y):
return self.clf.fit(x, y)
def feature_importances(self, x, y):
return self.clf.fit(x, y).feature_importances_
def get_oof(clf, x_train, y_train, x_test):
oof_train = np.zeros((ntrain,))
oof_test = np.zeros((ntest,))
oof_test_skf = np.empty((NFOLDS, ntest))
for i, (train_index, test_index) in enumerate(kf):
x_tr = x_train[train_index]
y_tr = y_train[train_index]
x_te = x_train[test_index]
clf.train(x_tr, y_tr)
oof_train[test_index] = clf.predict(x_te)
oof_test_skf[i, :] = clf.predict(x_test)
oof_test[:] = oof_test_skf.mean(axis=0)
return oof_train.reshape(-1, 1), oof_test.reshape(-1, 1)
现在让我们准备五个学习模型作为我们的第一级分类。这些模型都可以通过Sklearn库方便地调用,如下所示
1.Random Forest classifier
2.Extra Trees classifier
3.AdaBoost classifer
4.Gradient Boosting classifer
5.Support Vector Machine
输入上述分类器的参数
# 随机森林的参数
rf_params = {
'n_jobs': -1,
'n_estimators': 100,
'warm_start': True,
#'max_features': 0.2,
'max_depth': 6,
'min_samples_leaf': 2,
'max_features': 'sqrt',
'verbose': 0
}
# Extra Trees的参数
et_params = {
'n_jobs': -1,
'n_estimators': 100,
#'max_features': 0.5,
'max_depth': 8,
'min_samples_leaf': 2,
'verbose': 0
}
# AdaBoost的参数
ada_params = {
'n_estimators': 100,
'learning_rate': 0.01
}
# Gradient Boosting的参数
gb_params = {
'n_estimators': 100,
#'max_features': 0.2,
'max_depth': 5,
'min_samples_leaf': 2,
'verbose': 0
}
# Support Vector Classifier的参数
svc_params = {
'kernel': 'linear',
'C': 0.025
}
第一级分类器
# 通过前面定义的SklearnHelper类创建5个对象来表示5个学习模型
rf = SklearnHelper(clf=RandomForestClassifier, seed=SEED, params=rf_params)
et = SklearnHelper(clf=ExtraTreesClassifier, seed=SEED, params=et_params)
ada = SklearnHelper(clf=AdaBoostClassifier, seed=SEED, params=ada_params)
gb = SklearnHelper(clf=GradientBoostingClassifier, seed=SEED, params=gb_params)
svc = SklearnHelper(clf=SVC, seed=SEED, params=svc_params)
# 创建包含train、test的Numpy数组,以提供给我们的模型
y_train = train['Survived'].ravel()
train = train.drop(['Survived'], axis=1)
x_train = train.values
# test = test.drop(['Parch', 'Embarked', 'Has_Cabin', 'IsAlone'], axis=1)
x_test = test.values
#这些将会作为新的特征被使用
et_oof_train, et_oof_test = get_oof(et, x_train, y_train, x_test) # Extra Trees
rf_oof_train, rf_oof_test = get_oof(rf, x_train, y_train, x_test) # Random Forest
ada_oof_train, ada_oof_test = get_oof(ada, x_train, y_train, x_test) # AdaBoost
gb_oof_train, gb_oof_test = get_oof(gb, x_train, y_train, x_test) # Gradient Boost
svc_oof_train, svc_oof_test = get_oof(svc, x_train, y_train, x_test) # Support Vector Classifier
现在已经获得了我们的第一级预测,我们可以把它看作是一组新的特性,作为下一个分类器的训练数据。
查看各个特征对上述分类器的重要性
rf_features = rf.feature_importances(x_train, y_train)
et_features = et.feature_importances(x_train, y_train)
ada_features = ada.feature_importances(x_train, y_train)
gb_features = gb.feature_importances(x_train, y_train)
cols = train.columns.values
feature_dataframe = pd.DataFrame({'features': cols,
'Random Forest feature importances': rf_features,
'Extra Trees feature importances': et_features,
'AdaBoost feature importances': ada_features,
'Gradient Boost feature importances': gb_features})
feature_dataframe['mean'] = feature_dataframe.mean(axis=1) # axis = 1 computes the mean row-wise
print(feature_dataframe.head(11))
画图查看各个分类器的相关性
base_predictions_train = pd.DataFrame( {'RandomForest': rf_oof_train.ravel(),
'ExtraTrees': et_oof_train.ravel(),
'AdaBoost': ada_oof_train.ravel(),
'GradientBoost': gb_oof_train.ravel()
})
data = [
go.Heatmap(
z= base_predictions_train.astype(float).corr().values ,
x=base_predictions_train.columns.values,
y= base_predictions_train.columns.values,
colorscale='Viridis',
showscale=True,
reversescale = True
)
]
py.iplot(data, filename='labelled-heatmap')
,这些模型彼此之间的相关性越低,得分越高。
第二级分类器xgboost
x_train = np.concatenate((et_oof_train, rf_oof_train, ada_oof_train, gb_oof_train, svc_oof_train), axis=1)
x_test = np.concatenate((et_oof_test, rf_oof_test, ada_oof_test, gb_oof_test, svc_oof_test), axis=1)
gbm = xgb.XGBClassifier(
#learning_rate=0.01,
n_estimators=2000,
max_depth=4,
min_child_weight=2,
# gamma=1,
gamma=0.9,
subsample=0.8,
colsample_bytree=0.8,
objective='binary:logistic',
nthread=-1,
scale_pos_weight=1).fit(x_train, y_train)
predictions = gbm.predict(x_test)
提交
StackingSubmission = pd.DataFrame({'PassengerId': PassengerId,
'Survived': predictions})
StackingSubmission.to_csv("StackingSubmission.csv", index=False)
提交后的分数,排名如下
总结
相比于其他的kernels,这个kernels的特征工程方面做的不突出(但还是比我之前的好很多,哈哈哈),突出的方面是用了新的方法Stacking,这个其他人在Titanic比赛中没有用到过的,这也是他排第一的原因。
进一步改善的步骤
必须指出的是,上述步骤只是显示了一个非常简单的方法。听说过在Kaggle的最高级别比赛中创建的组合,其中包括stacked classifiers的巨大组合,以及超过2级的stacking级别。