机器学习 —— 分类预测与集成学习（下）

# merged_data = train_data.append(test_data)    # 合并训练集和测试集
# FutureWarning: The frame.append method is deprecated and will be removed from pandas in a future version. Use pandas.concat instead.
# 弃用警告，append方法将弃用，建议使用concat方法，而concat是pandas的函数，pandas.concat()通常用来连接DataFrame对象。要写成df=pd.concat([df1,df2])类似这样
# concat()通常用来连接DataFrame对象,默认情况下是对两个DataFrame对象进行纵向连接
merged_data = pd.concat([train_data, test_data]) # 合并训练集和测试集
for column in merged_data.columns:
    if merged_data[column].dtype == 'object':
        merged_data[column] = pd.Categorical(merged_data[column]).codes#Categorical函数，将文本转换成数值。

train_data = merged_data[0:train_data.shape[0]]
test_data = merged_data[train_data.shape[0]:]

print("训练数据维度：", train_data.shape)#.shape返回的是元组
print("测试数据维度：", test_data.shape)
print(test_data.head())#head()默认情况下,它会显示5行系列数据

X_train = train_data.iloc[:, :-1]
y_train = train_data['wage_class']

X_test = test_data.iloc[:, :-1]
y_test = test_data['wage_class']

cv_params = {'max_depth': [3, 5, 7], 'min_child_weight': [1, 3, 5]}
ind_params = {'learning_rate': 0.1, 'n_estimators': 1000, 'seed': 0,
              'subsample': 0.8, 'colsample_bytree': 0.8,
              'objective': 'binary:logistic'}

from xgboost import XGBClassifier#加载xgboost分类器（封装好的直接用）
from sklearn.model_selection import GridSearchCV#模型调参方法：GridSearch+CV（网格搜索+交叉验证）

print("训练模型并选择最优参数......")
# 使用5-fold cross-validation（5折交叉验证）来优选最佳的模型
optimized_GBM = GridSearchCV(XGBClassifier(**ind_params), cv_params, scoring='accuracy', cv=5, n_jobs=-1, verbose=10)#网格搜索：自动调参
# fit生成规则
optimized_GBM.fit(X_train, y_train)#训练

print("最佳参数：", optimized_GBM.best_params_)
# cv_results_返回模型训练过程所有详细信息  cv交叉验证
means = optimized_GBM.cv_results_['mean_test_score']#输出交叉验证结果 将规则应用于训练集
stds = optimized_GBM.cv_results_['std_test_score']#输出交叉验证结果 将规则应用于测试集

for mean, std, params in zip(means, stds, optimized_GBM.cv_results_['params']):    
    print("%0.5f (+/-%0.05f) for %r" % (mean, std * 2, params))#输出结果

from sklearn.metrics import classification_report
# predict()函数是Python中预测函数，常用于预测测试集数据，返回的是样本所属的类别标签。
y_pred = optimized_GBM.predict(X_test)#预测/测试
print(classification_report(y_test, y_pred))

cv_params = {'learning_rate': [0.1, 0.05, 0.01], 'subsample': [0.7, 0.8, 0.9]}
ind_params = {'max_depth': 3, 'n_estimators': 1000, 'seed': 0,'min_child_weight': 5, 'colsample_bytree': 0.8, 'objective': 'binary:logistic'}

print("训练模型并选择最优参数......")
# 使用5-fold cross-validation来优选最佳的模型
optimized_GBM = GridSearchCV(XGBClassifier(**ind_params), cv_params, scoring='accuracy', cv=5, n_jobs=-1, verbose=10)#网格搜索：自动调参
optimized_GBM.fit(X_train, y_train)#训练   # fit生成规则

print("最佳参数：", optimized_GBM.best_params_)
# cv_results_返回模型训练过程所有详细信息  cv交叉验证
means = optimized_GBM.cv_results_['mean_test_score']#输出交叉验证结果
stds = optimized_GBM.cv_results_['std_test_score']#输出交叉验证结果

for mean, std, params in zip(means, stds, optimized_GBM.cv_results_['params']):    
    print("%0.5f (+/-%0.05f) for %r" % (mean, std * 2, params))#输出结果

# 构建最优XGBoost模型
ind_params = {'max_depth': 3, 'min_child_weight': 5, 'learning_rate': 0.05, 'subsample': 0.8, 'n_estimators': 1000, 'seed': 0, 'colsample_bytree': 0.8, 'objective': 'binary:logistic'}
eval_set = [(X_test, y_test)]

model = XGBClassifier(**ind_params)
# 通过设置early_stopping_rounds可指定停止训练的时机。当测试集上的误差在early_stopping_rounds轮迭代之内都没有降低的话，就停止训练
result = model.fit(X_train, y_train, early_stopping_rounds=100, eval_metric="error", eval_set=eval_set, verbose=20)#训练
print("最佳迭代次数:", result.best_iteration)#通过best_iteration属性可获得最佳的迭代次数

# predict()函数是Python中预测函数，常用于预测测试集数据，返回的是样本所属的类别标签。
y_pred = model.predict(X_test, ntree_limit=result.best_iteration)#预测/测试
print(classification_report(y_test, y_pred))

import seaborn as sns

#绘制数据集中每两个特征(列)之间的相关性热力图
#步骤：创建画布→获取数据（train_data.corr()）→绘图→show
# 创建画布
# style must be one of white, dark, whitegrid, darkgrid, ticks
sns.set(style='white')
# 建立画布,figsize设置画布大小
plt.figure(figsize=(10, 8))
# 获取数据并绘制热力图
corr = train_data.corr()
# annot为True时，可设置各个参数，包括大小，颜色，加粗，斜体字等
# cmap：matplotlib的colormap名称或颜色对象
# fmt，格式设置
sns.heatmap(corr, annot=True, cmap='Greens',fmt=".2f")#heatmap将矩形数据绘制为颜色编码矩阵
# 保存图片
plt.savefig("three.png") 
# 显示图像
plt.show()

from xgboost import XGBClassifier
from sklearn.metrics import classification_report

# 指定超参数
ind_params = {'max_depth': 3, 'min_child_weight': 5, 'learning_rate': 0.05, 'subsample': 0.8, 'n_estimators': 1000, 'seed': 0, 'colsample_bytree': 0.8, 'objective': 'binary:logistic'}

# 去除强相关的列
# drop()函数的功能是通过指定的索引或标签名称,也就是行名称或者列名称进行删除数据。
X_train_reduced = X_train.drop(columns = ['education_num','relationship'])#drop删除某一列
X_test_reduced = X_test.drop(columns = ['education_num','relationship'])#drop删除某一列
eval_set = [(X_test_reduced, y_test)]

# 训练模型
model = XGBClassifier(**ind_params)
result = model.fit(X_train_reduced, y_train, early_stopping_rounds=100, eval_metric="error", eval_set=eval_set, verbose=50)#fit训练
print("最佳迭代次数:", result.best_iteration)

# 预测并计算性能
# predict()函数是Python中预测函数，常用于预测测试集数据，返回的是样本所属的类别标签。
y_pred = model.predict(X_test_reduced, ntree_limit=result.best_iteration)#预测/测试
print(classification_report(y_test, y_pred))

# 定义年龄区间
age_bins = [20, 30, 40, 50, 60, 70]    # 区间0：0~20，区间1：20~30，......区间6：70~

# 将'age'转换成区间
# digitize()主要用于将一组数据进行分区
X_train_reduced['age'] = np.digitize(X_train['age'], bins=age_bins)#返回每个值属于bins的索引
X_test_reduced['age'] = np.digitize(X_test['age'], bins=age_bins)#返回每个值属于bins的索引
print(X_train_reduced['age'].unique())#提取数据集合中的唯一值（去除重复的元素）
eval_set = [(X_test_reduced, y_test)]

# 指定超参数
ind_params = {'max_depth': 3, 'min_child_weight': 5, 'learning_rate': 0.05, 'subsample': 0.8, 'n_estimators': 1000, 'seed': 0, 'colsample_bytree': 0.8, 'objective': 'binary:logistic'}

# 训练模型
model = XGBClassifier(**ind_params)
result = model.fit(X_train_reduced, y_train, early_stopping_rounds=100, eval_metric="error", eval_set=eval_set, verbose=20)#fit训练
print("最佳迭代次数:", result.best_iteration)

# 预测并计算性能
# predict()函数是Python中预测函数，常用于预测测试集数据，返回的是样本所属的类别标签。
y_pred = model.predict(X_test_reduced, ntree_limit=result.best_iteration)#测试/预测
print(classification_report(y_test, y_pred))