import pandas as pd
import lightgbm as lgb
import argparse

def feature_importance_rank_and_save(train_csv_path, test_csv_path, n_estimators,
                                    output_train_path=None, output_test_path=None):
    """
    基于LightGBM计算特征重要性，排序后重新保存训练集和测试集
    :param train_csv_path: 训练集CSV路径
    :param test_csv_path: 测试集CSV路径
    :param n_estimators: 树的数量
    :param output_train_path: 输出排序后的训练集路径
    :param output_test_path: 输出排序后的测试集路径
    """
    # ===================== 1. 读取数据 =====================
    print("正在读取数据...")
    train_df = pd.read_csv(train_csv_path)
    test_df = pd.read_csv(test_csv_path)

    # 数据格式校验
    if train_df.shape[1] < 3:
        raise ValueError("数据格式错误：至少需要ID列、标签列、1个特征列")

    # 拆分ID、标签、特征（严格按照需求：第1列ID，第2列标签，第3列开始为特征）
    id_col = train_df.columns[0]    # ID列名
    label_col = train_df.columns[1]  # 标签列名
    feature_cols = train_df.columns[2:]  # 特征列名

    X_train = train_df[feature_cols]
    y_train = train_df[label_col]

    # ===================== 2. 训练LightGBM模型 =====================
    print(f"正在训练LightGBM模型，n_estimators = {n_estimators} ...")
    # 基础参数（自动判断分类/回归）
    params = {
        "boosting_type": "gbdt",
        "objective": "binary" if len(y_train.unique()) <= 2 else "regression",
        "metric": "auc",
        "random_state": 42,
        "verbose": -1,
        "n_estimators": n_estimators  # 传入命令行参数
    }

    # 初始化并训练模型
    if params["objective"] == "binary":
        lgb_model = lgb.LGBMClassifier(**params)
    else:
        lgb_model = lgb.LGBMRegressor(**params)

    lgb_model.fit(X_train, y_train)

    # ===================== 3. 计算并排序特征重要性 =====================
    print("正在排序特征重要性...")
    feature_importance = pd.DataFrame({
        "feature": feature_cols,
        "importance": lgb_model.feature_importances_
    })

    # 按重要性降序（重要→不重要）
    feature_importance_sorted = feature_importance.sort_values(by="importance", ascending=False).reset_index(drop=True)
    feature_importance_sorted["rank"] = range(1, len(feature_importance_sorted) + 1)

    # 打印排名
    print("\n===== 特征重要性排名（重要→不重要）=====")
    print(feature_importance_sorted.to_string(index=False))

    # 排序后的特征列表
    sorted_feature_cols = feature_importance_sorted["feature"].tolist()

    # ===================== 4. 自动生成输出文件名 =====================
    if output_train_path is None:
        output_train_path = f"训练集_SFbyLGBM_{n_estimators}.csv"
    if output_test_path is None:
        output_test_path = f"测试集_SFbyLGBM_{n_estimators}.csv"

    # ===================== 5. 重新保存数据集 =====================
    print("\n正在保存排序后的数据集...")
    new_columns = [id_col, label_col] + sorted_feature_cols

    ranked_train_df = train_df[new_columns]
    ranked_test_df = test_df[new_columns]

    # 保存CSV（Excel不乱码）
    ranked_train_df.to_csv(output_train_path, index=False, encoding="utf-8-sig")
    ranked_test_df.to_csv(output_test_path, index=False, encoding="utf-8-sig")

    print(f"\n✅ 排序完成！")
    print(f"📄 训练集已保存：{output_train_path}")
    print(f"📄 测试集已保存：{output_test_path}")

if __name__ == "__main__":
    # ===================== argparse 配置 =====================
    parser = argparse.ArgumentParser(description="LightGBM 特征重要性排序工具")
    parser.add_argument("--train", required=True, help="训练集CSV文件路径")
    parser.add_argument("--test", required=True, help="测试集CSV文件路径")
    parser.add_argument("--n_estimators", type=int, required=True, help="LightGBM 树的数量 n_estimators")
    
    args = parser.parse_args()

    # 运行主函数
    feature_importance_rank_and_save(
        train_csv_path=args.train,
        test_csv_path=args.test,
        n_estimators=args.n_estimators
    )