# -*- coding: utf-8 -*-

import pandas as pd
from flaml import AutoML
import mlflow
import mlflow.sklearn
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score, roc_auc_score

# 1. 准备数据（示例）
# df = pd.read_csv("financial_data.csv")
# X = df.drop("default", axis=1)
# y = df["default"]

# 为演示，用 sklearn 的 make_classification 生成模拟金融数据
from sklearn.datasets import make_classification
X, y = make_classification(n_samples=10000, n_features=20, n_informative=10, 
                           n_redundant=5, n_clusters_per_class=1, random_state=42)

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# 2. 启动 MLflow 实验
mlflow.set_experiment("FLAML_cc1")

with mlflow.start_run(run_name="FLAML_cc1"):

    # 3. 配置 FLAML
    automl = AutoML()
    settings = {
        "time_budget": 60,          # 60秒自动搜索
        "metric": "roc_auc",        # 金融常用 AUC
        "task": "classification",
        "log_file_name": "flaml.log",
        "seed": 42,
        "estimator_list": ["lgbm", "xgboost", "rf"],  # 限定金融常用模型
    }

    # 4. 开始自动训练
    automl.fit(X_train, y_train, **settings)

    # 5. 获取最佳模型和预测
    y_pred_proba = automl.predict_proba(X_test)[:, 1]
    test_auc = roc_auc_score(y_test, y_pred_proba)
    test_acc = accuracy_score(y_test, automl.predict(X_test))

    # 6. 记录到 MLflow
    # (a) 记录 FLAML 自动选择的超参数
    mlflow.log_params(automl.best_config)

    # (b) 记录评估指标
    mlflow.log_metric("test_roc_auc", test_auc)
    mlflow.log_metric("test_accuracy", test_acc)

    # (c) 记录使用的模型类型
    mlflow.log_param("best_estimator", automl.best_estimator)

    # (d) 保存特征数量（金融建模常关注特征维度）
    mlflow.log_param("n_features", X_train.shape[1])

    # (e) 保存最终模型（FLAML 的 best_model 是 sklearn 兼容模型）
    mlflow.sklearn.log_model(automl.model, "flaml_best_model")

    print(f"Best model: {automl.best_estimator}")
    print(f"Test AUC: {test_auc:.4f}")
    if hasattr(automl.model, "feature_importances_"):
        fi = automl.model.feature_importances_
        for i, imp in enumerate(fi):
            mlflow.log_metric(f"feature_{i}_importance", imp)
        

冰与火的战歌：Windows内核攻防实战高级班！从零到实战，融合AI与Windows内核攻防全技术栈，打造具备自动化能力的内核开发高手。