batch_runner_all.py 8.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256
  1. #!/usr/bin/env python3
  2. """
  3. 批量运行器 - 批量执行Complete Agent Flow
  4. ========================================
  5. 此脚本可以批量运行多次完整的智能体工作流,每次运行会创建独立的日志文件夹。
  6. 使用方法:
  7. python batch_runner.py
  8. 配置参数:
  9. - 运行次数: RUNS = 10
  10. - 行业: INDUSTRY = "农业"
  11. - 数据文件: DATA_FILE = "data_files/交易流水样例数据.csv"
  12. - 查询问题: QUESTION = "请生成一份详细的农业经营贷流水分析报告..."
  13. 文件夹结构:
  14. api_results_1/ # 第一次运行的日志
  15. api_results_2/ # 第二次运行的日志
  16. ...
  17. api_results_10/ # 第十次运行的日志
  18. 作者: Big Agent Team
  19. 版本: 1.0.0
  20. 创建时间: 2024-12-20
  21. """
  22. import asyncio
  23. import os
  24. from datetime import datetime
  25. from typing import List, Dict, Any
  26. import sys
  27. import os
  28. # 添加项目根目录到路径,以便导入config
  29. current_dir = os.path.dirname(os.path.abspath(__file__))
  30. parent_dir = os.path.dirname(current_dir)
  31. sys.path.insert(0, parent_dir)
  32. os.environ["LANGCHAIN_TRACING_V2"] = "false"
  33. os.environ["LANGCHAIN_API_KEY"] = ""
  34. # 禁用 LangGraph 的追踪
  35. os.environ["LANGSMITH_TRACING"] = "false"
  36. # 根据执行方式选择导入方式
  37. if __name__ == "__main__":
  38. # 直接执行文件时,使用绝对导入
  39. from llmops.complete_agent_flow_rule import run_complete_agent_flow
  40. from llmops.agents.data_manager import DataManager
  41. else:
  42. # 作为模块导入时,使用相对导入
  43. from .complete_agent_flow_rule import run_complete_agent_flow
  44. from .agents.data_manager import DataManager
  45. import config
  46. # ========== 配置参数 ==========
  47. RUNS = 10 # 运行次数
  48. INDUSTRY = "农业" # 行业
  49. ORIGINAL_DATA_FILE="data_files/11111.png" # 原始测试数据文件PDF
  50. DATA_FILE = "data_files/交易流水样例数据.csv" # 数据文件路径
  51. QUESTION = "请生成一份详细的农业经营贷流水分析报告,需要包含:1.总收入和总支出统计 2.收入笔数和支出笔数 3.各类型收入支出占比分析 4.交易对手收入支出TOP3排名 5.按月份的收入支出趋势分析 6.账户数量和交易时间范围统计 7.资金流入流出月度统计等全面指标" # 分析查询
  52. # ==============================
  53. async def run_single_flow(run_id: str, question: str, industry: str, data: List[Dict[str, Any]], file_name: str) -> Dict[str, Any]:
  54. """
  55. 运行单个工作流实例
  56. Args:
  57. run_id: 运行ID
  58. question: 用户查询
  59. industry: 行业
  60. data: 数据集
  61. file_name: 文件名
  62. Returns:
  63. 运行结果
  64. """
  65. print(f"\n{'='*60}")
  66. print(f"🚀 开始运行 #{run_id}")
  67. print(f"📁 日志文件夹: api_results_{run_id}")
  68. print(f"{'='*60}")
  69. # 设置环境变量,让所有agent使用正确的文件夹
  70. os.environ['FLOW_RUN_ID'] = run_id
  71. try:
  72. result = await run_complete_agent_flow(
  73. question=question,
  74. industry=industry,
  75. data=data,
  76. file_name=file_name,
  77. api_key=config.DEEPSEEK_API_KEY,
  78. session_id=f"batch-run-{run_id}"
  79. )
  80. if result.get('success'):
  81. summary = result.get('execution_summary', {})
  82. print(f"✅ 运行 #{run_id} 成功完成")
  83. print(f" 规划步骤: {summary.get('planning_steps', 0)}")
  84. print(f" 指标计算: {summary.get('metrics_computed', 0)}")
  85. else:
  86. print(f"❌ 运行 #{run_id} 失败: {result.get('error', '未知错误')}")
  87. return result
  88. except Exception as e:
  89. print(f"❌ 运行 #{run_id} 发生异常: {e}")
  90. return {
  91. "success": False,
  92. "error": str(e),
  93. "run_id": run_id
  94. }
  95. async def data_standardize():
  96. """
  97. 提取数据,进行标准化处理
  98. """
  99. from llmops.agents.data_stardard import TransactionParserAgent
  100. from llmops.config import DEEPSEEK_API_KEY, multimodal_api_url
  101. # 数据标准化agent
  102. standard_agent = TransactionParserAgent(
  103. api_key=DEEPSEEK_API_KEY,
  104. multimodal_api_url=multimodal_api_url
  105. )
  106. # 执行解析
  107. full_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", ORIGINAL_DATA_FILE)
  108. try:
  109. result = await standard_agent.run_workflow_task(full_path)
  110. if result["status"] == "success":
  111. print(f"🎯 数据标准化任务完成!")
  112. # 标准化后的文件
  113. standard_file_path = result['file_path']
  114. return standard_file_path
  115. else:
  116. print(f"❌ 数据标准化任务失败: {result['message']}")
  117. raise ValueError("数据标准化异常")
  118. except Exception as e:
  119. print(f"数据标准化处理异常:{e}")
  120. raise e
  121. async def run_batch(runs: int, question: str, industry: str, data_file: str):
  122. """
  123. 批量运行工作流
  124. Args:
  125. runs: 运行次数
  126. question: 用户查询
  127. industry: 行业
  128. data_file: 原始数据文件路径,格式支持pdf/img/csv
  129. """
  130. print("🚀 批量运行器启动")
  131. print(f"📊 计划运行次数: {runs}")
  132. print(f"🏭 行业: {industry}")
  133. print(f"📁 数据文件: {data_file}")
  134. print(f"❓ 查询: {question}")
  135. print(f"{'='*80}")
  136. # 检查API密钥
  137. if not config.DEEPSEEK_API_KEY:
  138. print("❌ 未找到API密钥,请检查config.py")
  139. return
  140. # 运行结果统计
  141. successful_runs = 0
  142. failed_runs = 0
  143. results = []
  144. # 运行总时长,秒
  145. total_time = 0
  146. import time
  147. # 逐个运行
  148. for i in range(1, runs + 1):
  149. run_id = str(i)
  150. # 设置环境变量,让所有agent使用正确的文件夹
  151. os.environ['FLOW_RUN_ID'] = run_id
  152. start_time = time.perf_counter()
  153. try:
  154. # 执行数据标准化
  155. standard_file_path = await data_standardize()
  156. print(f"标准化后文件:{standard_file_path}")
  157. data_file = standard_file_path
  158. # 标准化数据加载
  159. data = DataManager.load_data_from_csv_file(data_file)
  160. print(f"📊 数据加载成功: {len(data)} 条记录")
  161. except Exception as e:
  162. print(f"❌ 数据加载失败: {e}")
  163. return
  164. # 单次执行
  165. result = await run_single_flow(run_id, question, industry, data, os.path.basename(data_file))
  166. end_time = time.perf_counter()
  167. total_time += (end_time - start_time)
  168. results.append(result)
  169. if result.get('success'):
  170. successful_runs += 1
  171. else:
  172. failed_runs += 1
  173. # 添加短暂延迟,避免API调用过于频繁
  174. if i < runs: # 最后一次不需要延迟
  175. await asyncio.sleep(1)
  176. # 输出统计结果
  177. print(f"\n{'='*80}")
  178. print("📊 批量运行完成统计")
  179. print(f"{'='*80}")
  180. print(f"总运行次数: {runs}")
  181. print(f"总运行总用时: {total_time:.2f}秒,单次用时:{total_time/runs:.2f}秒")
  182. print(f"成功次数: {successful_runs}")
  183. print(f"失败次数: {failed_runs}")
  184. print(f"成功率: {successful_runs/runs*100:.1f}%")
  185. # 显示各运行的日志文件夹
  186. print(f"\n📁 日志文件夹列表:")
  187. for i in range(1, runs + 1):
  188. folder_name = f"api_results_{i}"
  189. status = "✅" if results[i-1].get('success') else "❌"
  190. print(f" {status} {folder_name}")
  191. print("\n🎉 批量运行完成!")
  192. print(f"💡 提示: 每次运行的完整日志保存在对应的 api_results_[数字] 文件夹中")
  193. def main():
  194. """主函数"""
  195. print("🚀 使用配置参数运行批量任务")
  196. print(f"📊 运行次数: {RUNS}")
  197. print(f"🏭 行业: {INDUSTRY}")
  198. print(f"📁 数据文件: {ORIGINAL_DATA_FILE}")
  199. print(f"❓ 查询: {QUESTION[:50]}...")
  200. print("-" * 80)
  201. # 运行批量任务
  202. asyncio.run(run_batch(
  203. runs=RUNS,
  204. question=QUESTION,
  205. industry=INDUSTRY,
  206. data_file=ORIGINAL_DATA_FILE
  207. ))
  208. if __name__ == "__main__":
  209. main()