jiangcang_vj/result_validate.py

import os
import datetime
import pandas as pd
from data_loader import mongo_con_parse, validate_one_line, fill_hourly_crawl_date


def validate_process(node, date, pred_time_str):

    output_dir = f"./validate/{node}_{date}"
    os.makedirs(output_dir, exist_ok=True)

    object_dir = "./data_shards"
    csv_file = f'future_predictions_{pred_time_str}.csv'  
    csv_path = os.path.join(object_dir, csv_file)

    try:
        df_predict = pd.read_csv(csv_path)
    except Exception as e:
        print(f"read {csv_path} error: {str(e)}")
        df_predict = pd.DataFrame()
    
    if df_predict.empty:
        print(f"预测数据为空")
        return

    # fly_day = df_predict['flight_day'].unique()[0]

    client, db = mongo_con_parse()

    count = 0
    for idx, row in df_predict.iterrows(): 
        city_pair = row['city_pair']
        flight_day = row['flight_day']
        flight_number_1 = row['flight_number_1']
        flight_number_2 = row['flight_number_2']
        baggage = row['baggage']
        valid_begin_hour = row['valid_begin_hour'] 
        df_val= validate_one_line(db, city_pair, flight_day, flight_number_1, flight_number_2, baggage, valid_begin_hour)
        # 有可能在当前验证时刻，数据库里没有在valid_begin_hour之后的数据
        if not df_val.empty:
            df_val_f = fill_hourly_crawl_date(df_val, rear_fill=2)
            df_val_f = df_val_f[df_val_f['is_filled']==0]    # 只要原始数据，不要补齐的
            if df_val_f.empty:
                drop_flag = 0
                first_drop_amount = pd.NA
                first_drop_hours = pd.NA
                last_hours_util = pd.NA
                last_update_hour = pd.NA
                list_change_price = []
                list_change_hours = []
            else:
                # 有效数据的最后一行
                last_row = df_val_f.iloc[-1]
                last_hours_util = last_row['hours_until_departure']
                last_update_hour = last_row['update_hour']
                
                # 价格变化过滤
                df_price_changes = df_val_f.loc[
                    df_val_f["adult_total_price"].shift() != df_val_f["adult_total_price"]
                ].copy()
            
                # 价格变化幅度
                df_price_changes['change_amount'] = df_price_changes['adult_total_price'].diff().fillna(0)

                # 找到第一个 change_amount 小于 -10 的行
                first_negative_change = df_price_changes[df_price_changes['change_amount'] < -10].head(1)

                # 提取所需的值
                if not first_negative_change.empty:
                    drop_flag = 1
                    first_drop_amount = first_negative_change['change_amount'].iloc[0].round(2)
                    first_drop_hours = first_negative_change['hours_until_departure'].iloc[0]
                else:
                    drop_flag = 0
                    first_drop_amount = pd.NA
                    first_drop_hours = pd.NA
                    
                list_change_price = df_price_changes['adult_total_price'].tolist()
                list_change_hours = df_price_changes['hours_until_departure'].tolist()
        
        else:
            drop_flag = 0
            first_drop_amount = pd.NA
            first_drop_hours = pd.NA
            last_hours_util = pd.NA
            last_update_hour = pd.NA
            list_change_price = []
            list_change_hours = []

        safe_sep = "; "
        
        df_predict.at[idx, 'change_prices'] = safe_sep.join(map(str, list_change_price))
        df_predict.at[idx, 'change_hours'] = safe_sep.join(map(str, list_change_hours))
        df_predict.at[idx, 'last_hours_util'] = last_hours_util
        df_predict.at[idx, 'last_update_hour'] = last_update_hour
        df_predict.at[idx, 'first_drop_amount'] = first_drop_amount * -1  # 负数转正数
        df_predict.at[idx, 'first_drop_hours'] = first_drop_hours
        df_predict.at[idx, 'drop_flag'] = drop_flag

        count += 1
        if count % 5 == 0:
            print(f"cal count: {count}")
    
    print(f"计算结束")
    client.close()

    timestamp_str = datetime.datetime.now().strftime("%Y%m%d%H%M%S")
    save_scv = f"result_validate_{node}_{pred_time_str}_{timestamp_str}.csv"
    
    output_path = os.path.join(output_dir, save_scv)
    df_predict.to_csv(output_path, index=False, encoding="utf-8-sig")
    print(f"保存完成: {output_path}")


if __name__ == "__main__":
    node, date, pred_time_str = "node0108", "0109", "202601091100"
    validate_process(node, date, pred_time_str)