使用mlflow run运行项目时，如何让Python脚本指定的experiment-id生效？

我用MLflow在Python脚本中跟踪运行记录，脚本内已编写实验检查与指定逻辑，确保运行记录归属目标实验：

mlflow.set_tracking_uri("http://127.0.0.1:5000")
exps = mlflow.search_experiments()
exp_names = [e.name for e in exps]
if EXP_NAME in exp_names:
   experiment_id = exps[exp_names.index(EXP_NAME)].experiment_id
else:
   experiment_id = mlflow.create_experiment(EXP_NAME)
print(f"Created New Experiment > ID: {experiment_id}")
experiment = mlflow.set_experiment(experiment_id=experiment_id)
with mlflow.start_run(run_name="fantastic-run", experiment_id=experiment_id):
    # 运行逻辑
    mlflow.endrun()

直接用python my_python_script.py执行脚本一切正常，但通过MLproject的my-entry条目，用mlflow run -e my-entry path_to_my_mlflow_project/执行时，系统会默认使用ID=0的实验，且脚本抛出错误：

mlflow.exceptions.MlflowException: Cannot start run with ID 4552f1852c0d4eafae72b84d38af7346 because active run ID does not match environment run ID. Make sure --experiment-name or --experiment-id matches experiment set with set_experiment(), or just use command-line arguments

ERROR mlflow.cli: === Run (ID '4552f1852c0d4eafae72b84d38af7346') failed ===

解决方法 #

• 命令行直接指定实验
  在mlflow run命令中添加--experiment-name或--experiment-id参数，与脚本内的EXP_NAME保持一致：

  mlflow run -e my-entry path_to_my_mlflow_project/ --experiment-name "你的实验名称"
  

  这样MLflow会直接使用指定实验，避免脚本内的实验设置与命令行启动的运行冲突。

• 修改脚本适配MLproject运行场景
  用mlflow run启动时，MLflow会自动创建活跃运行，此时脚本手动调用mlflow.start_run会触发冲突。可以先检查是否存在活跃运行，再决定是否创建新运行：

  mlflow.set_tracking_uri("http://127.0.0.1:5000")
  exps = mlflow.search_experiments()
  exp_names = [e.name for e in exps]
  if EXP_NAME in exp_names:
     experiment_id = exps[exp_names.index(EXP_NAME)].experiment_id
  else:
     experiment_id = mlflow.create_experiment(EXP_NAME)
  print(f"Created New Experiment > ID: {experiment_id}")
  
  # 检查当前是否有活跃运行，无则启动新运行，有则复用
  active_run = mlflow.active_run()
  if not active_run:
      with mlflow.start_run(run_name="fantastic-run", experiment_id=experiment_id):
          # 你的运行逻辑代码
          pass
  else:
      with mlflow.start_run(run_id=active_run.info.run_id, experiment_id=experiment_id):
          # 你的运行逻辑代码
          pass
  

• 在MLproject文件中配置实验参数
  可以在MLproject的entry点中定义实验参数，让脚本读取命令行传入的值：
  首先修改MLproject文件：

  name: my-project
  entry_points:
    my-entry:
      command: "python my_python_script.py --experiment-name {experiment_name}"
      parameters:
        experiment_name:
          type: string
          default: "默认实验名称"
  

  然后修改脚本，添加参数解析：

  import argparse
  parser = argparse.ArgumentParser()
  parser.add_argument("--experiment-name", type=str, default="默认实验名称")
  args = parser.parse_args()
  EXP_NAME = args.experiment_name
  
  # 原有的实验检查逻辑保持不变
  mlflow.set_tracking_uri("http://127.0.0.1:5000")
  exps = mlflow.search_experiments()
  exp_names = [e.name for e in exps]
  if EXP_NAME in exp_names:
     experiment_id = exps[exp_names.index(EXP_NAME)].experiment_id
  else:
     experiment_id = mlflow.create_experiment(EXP_NAME)
  print(f"Created New Experiment > ID: {experiment_id}")
  
  # 复用活跃运行或启动新运行的逻辑同上
  

  运行时通过-P传递实验名称：

  mlflow run -e my-entry path_to_my_mlflow_project/ -P experiment-name="你的实验名称"
  

内容的提问来源于stack exchange，提问作者inarighas