实验二 泰坦尼克号获救人员预测

import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.preprocessing import LabelEncoder
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import GridSearchCV

data_raw = pd.read_csv('train.csv')
data_val = pd.read_csv('test.csv')
# 列名转换成小写
# data_raw.columns=data_raw.columns.str.lower()
#print(data_raw.head())
#print(data_raw.info())
# 统计获救人员情况
#print(data_raw['Survived'].value_counts())
# 绘图统计
sns.countplot(data_raw['Survived'])
plt.show()
# 查看数据集中的空值
#print(data_raw.isnull().sum())
#print(data_val.isnull().sum())
# 数据清洗
#print(data_raw.describe())
# 补足年龄、票价和登船港口空缺值，年龄和票价用数据的中位数
data_raw['Age'].fillna(data_raw['Age'].median(), inplace=True)
data_val['Age'].fillna(data_val['Age'].median(), inplace=True)
data_raw['Fare'].fillna(data_raw['Fare'].median(), inplace=True)
data_val['Fare'].fillna(data_val['Fare'].median(), inplace=True)
# mode取embarked列出现频率最高的值，返回是一个Series
data_raw['Embarked'].fillna(data_raw['Embarked'].mode()[0], inplace=True)
data_val['Embarked'].fillna(data_val['Embarked'].mode()[0], inplace=True)
#print(data_raw.isnull().sum())
#print(data_val.isnull().sum())
# 删除没有用的列
data_raw.drop(['PassengerId', 'Cabin'], axis=1, inplace=True)
data_val.drop(['PassengerId', 'Cabin'], axis=1, inplace=True)
print(data_raw.info())
# 可选部分
# 构建新的特征：家庭成员的数量
data_raw['family_size'] = data_raw['SibSp'] + data_raw['Parch'] + 1
data_val['family_size'] = data_val['SibSp'] + data_val['Parch'] + 1
print(data_raw.info())
# 提取姓名中的称谓
# DataFrame中的apply方法就是将函数应用到由列或行形成的一维数组上
data_raw['title'] = data_raw['Name'].apply(lambda x:
                                           x.split(', ')[1]).apply(lambda x: x.split('. ')[0])
data_val['title'] = data_val['Name'].apply(lambda x:
                                           x.split(', ')[1]).apply(lambda x: x.split('. ')[0])
print(data_raw['title'].value_counts())
# 离散化票价
# qcut根据值的频率进行分组来选择箱子的的均匀间隔，即每个箱子中含有的数的数量相同（这里就是每组的元素个数一致）
data_raw['fare_bin'] = pd.qcut(data_raw['Fare'], 4)
data_val['fare_bin'] = pd.qcut(data_val['Fare'], 4)
print(data_raw['fare_bin'].value_counts())
# 离散化年龄
data_raw['age_bin'] = pd.cut(data_raw['Age'], 5)
data_val['age_bin'] = pd.cut(data_val['Age'], 5)
print(data_raw['age_bin'].value_counts())
# 基于LabelEncoder构建新字段
# 创建新字段type_code,值是LabelEncoder对type列进行one-hot编码
label = LabelEncoder()
data_raw['sex_code'] = label.fit_transform(data_raw['Sex'])
data_val['sex_code'] = label.fit_transform(data_val['Sex'])
print(data_raw['sex_code'])
data_raw['embarked_code'] = label.fit_transform(data_raw['Embarked'])
data_val['embarked_code'] = label.fit_transform(data_val['Embarked'])
data_raw['fare_bin_code'] = label.fit_transform(data_raw['fare_bin'])
data_val['fare_bin_code'] = label.fit_transform(data_val['fare_bin'])
data_raw['age_bin_code'] = label.fit_transform(data_raw['age_bin'])
data_val['age_bin_code'] = label.fit_transform(data_val['age_bin'])
data_raw['title_code'] = label.fit_transform(data_raw['title'])
data_val['title_code'] = label.fit_transform(data_val['title'])
# pd.set_option('max_columns', 20)
pd.options.display.max_columns = 20
print(data_raw.head())
print(data_raw.columns)
# 特征选择
# 目标集的列名
Target = ['Survived']
# 特征的列名
data_columns = ['Pclass', 'sex_code', 'age_bin_code', 'fare_bin_code',
                'embarked_code', 'family_size', 'title_code']
# 分割训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(
    data_raw[data_columns], data_raw[Target], test_size=0.3, random_state=5)
# max_features:每个决策树的最大特征数 n_jobs:多线程进行训练
# random_state如果希望可以重现，固定随机数种子。随机森林本质就是随机的，设置随机数种子可以彻底改变构建的模型
# 不设置每次构建的模型不同
# 实例化随机森林算法对象
rfc = RandomForestClassifier(max_features='auto', random_state=1, n_jobs=-1)
param_grid = {
    'n_estimators': [50, 100, 400, 700, 1000],  # 森林里的树木数量
    'criterion': ['gini', 'entropy'],  # 分割特征的测量方法
    'max_depth': [5, 8, 15, 25, 30],  # 树的最大深度
    'min_samples_split': [2, 4, 10, 12, 16],  # 节点划分最少样本数
    'min_samples_leaf': [1, 5, 10]  # 叶子节点的最小样本数
}
# 创建网络搜索对象，5折交叉验证，评估标准：scoring
gs = GridSearchCV(estimator=rfc, param_grid=param_grid, cv=5, scoring='accuracy', n_jobs=-1)
# # 训练模型
#gs.fit(X_train, y_train)
# # 平均验证精度最高分
#print(gs.best_score_)
# # 网格搜索后的最优参数
#print(gs.best_params_)
# 使用最佳参数构建随机森林
#y_train = y_train.values.ravel()
rfc = RandomForestClassifier(max_features='auto', random_state=1, n_jobs=-1, criterion='entropy',
                             max_depth=18, min_samples_leaf=5, min_samples_split=2, n_estimators=700)
rfc.fit(X_train, y_train)
score = rfc.score(X_test, y_test)
print(score)
y_predict = rfc.predict(X_test)

# 查看精确率、召回率，F1score
from sklearn.metrics import classification_report

report = classification_report(y_test, y_predict)
print(report)
# 查看roc_auc指标
from sklearn.metrics import roc_auc_score

roc = roc_auc_score(y_test, y_predict)
print(roc)

# 预测test.csv中的数据，特征需要和训练集选择的特征一致
y_predict = rfc.predict(data_val[data_columns])
print(y_predict)
data_val = pd.read_csv('test.csv')
# 新增一列预测结果
data_val['Survived'] = y_predict
# 导出excel
#data_val.to_excel('test1.xlsx')