#建模调参流程与逻辑回归

X_train = train_df.drop('Attrition', axis=1)
y_train = train_df['Attrition']
X_test = test_df

# 识别分类变量和数值变量
categorical_cols = X_train.select_dtypes(include=['object']).columns.tolist()
numerical_cols = X_train.select_dtypes(exclude=['object']).columns.tolist()

# 创建预处理管道
preprocessor = ColumnTransformer(
    transformers=[
        ('num', 'passthrough', numerical_cols),
        ('cat', OrdinalEncoder(), categorical_cols)
    ])

# 创建逻辑回归模型
logreg = LogisticRegression(max_iter=10000000)

# 创建包含预处理和模型的管道
clf = Pipeline(steps=[
    ('preprocessor', preprocessor),
    ('classifier', logreg)
])

# 定义超参数网格
param_grid = [
    {'classifier__C': [0.01, 0.1, 1, 10], 'classifier__penalty': ['l2'], 'classifier__solver': ['liblinear', 'lbfgs', 'sag', 'saga']},
    {'classifier__C': [0.01, 0.1, 1, 10], 'classifier__penalty': ['l1'], 'classifier__solver': ['liblinear', 'saga']}
]

# 创建 GridSearchCV 对象
grid_search = GridSearchCV(clf, param_grid, cv=5, scoring='roc_auc', n_jobs=-1)

# 训练模型
grid_search.fit(X_train, y_train)

# 输出最佳参数
best_params = grid_search.best_params_
best_score = grid_search.best_score_
print("最佳参数:", best_params)
print("最佳交叉验证得分:\nROC AUC:", best_score)

# 使用最佳模型进行预测
best_clf = grid_search.best_estimator_
y_pred_train = best_clf.predict(X_train)
y_pred_train_proba = best_clf.predict_proba(X_train)[:, 1]  # 获取训练集的预测概率

# 计算并输出训练集的评估结果
roc_auc = roc_auc_score(y_train, y_pred_train_proba)
print("训练集评估结果:")
print(f"ROC AUC: {roc_auc}")
print("分类报告:")
print(classification_report(y_train, y_pred_train))
print("混淆矩阵:")
print(confusion_matrix(y_train, y_pred_train))

# 自定义分类变量和数值变量
numerical_cols = ['Age', 'DailyRate', 'DistanceFromHome', 'HourlyRate', 'MonthlyIncome', 'MonthlyRate',
                  'NumCompaniesWorked', 'PercentSalaryHike', 'TotalWorkingYears', 'TrainingTimesLastYear',
                  'YearsAtCompany', 'YearsInCurrentRole', 'YearsSinceLastPromotion', 'YearsWithCurrManager']

categorical_cols = ['BusinessTravel', 'Department', 'Education', 'EducationField',
                    'EnvironmentSatisfaction', 'Gender', 'JobInvolvement', 'JobLevel', 'JobRole',
                    'JobSatisfaction', 'MaritalStatus', 'OverTime', 'PerformanceRating',
                    'RelationshipSatisfaction', 'StockOptionLevel', 'WorkLifeBalance']

# 创建预处理管道
preprocessor = ColumnTransformer(
    transformers=[
        ('num', StandardScaler(), numerical_cols),
        ('cat', OneHotEncoder(handle_unknown='ignore'), categorical_cols)
    ])

# 其余代码不变
# ...

# 这里切换成特征工程后的数据集
X_train = fe_train_df.drop('Attrition', axis=1)
y_train = fe_train_df['Attrition']

# 其余代码不变
# ...

# ...

# 定义超参数网格
param_grid = [
    {'classifier__C': [0.2, 0.25, 0.3, 0.35, 0.4], 'classifier__penalty': ['l2'], 'classifier__solver': ['liblinear', 'lbfgs', 'sag', 'saga']},
    {'classifier__C': [0.2, 0.25, 0.3, 0.35, 0.4], 'classifier__penalty': ['l1'], 'classifier__solver': ['liblinear', 'saga']}
]

# ...

# 使用特征工程后的测试集
X_test = fe_test_df

# 使用最佳模型进行预测
best_clf = grid_search.best_estimator_
y_pred_test_proba = best_clf.predict_proba(X_test)[:, 1]

# 将预测结果保存为提交文件
submission = pd.DataFrame({
        'id': test_ids,  # 使用之前保存的 id 列
        'Attrition': y_pred_test_proba
    })
submission.to_csv("logreg_submission.csv", index=False)

# 提交到 Kaggle
!kaggle competitions submit -c playground-series-s3e3 -f logreg_submission.csv -m "logreg_best"

# 查看提交结果
!kaggle competitions submissions -c playground-series-s3e3