TFX 1.0.0用Kubeflow编排流水线时报get_default_kubeflow_metadata_config属性缺失错误

AttributeError: module 'tfx.orchestration.experimental' has no attribute 'get_default_kubeflow_metadata_config'错误的核心原因是：TFX 1.0.0正式版已将Kubeflow编排相关能力从实验模块迁移到正式的tfx.orchestration.kubeflow模块，原experimental下的相关API已被移除。

修复步骤 #

• 调整导入语句，替换原experimental模块导入为Kubeflow正式模块

from tfx.orchestration.kubeflow import kubeflow_dag_runner

• 替换所有experimental前缀的API调用，对应调整为kubeflow_dag_runner下的方法：
  • 原experimental.get_default_kubeflow_metadata_config替换为kubeflow_dag_runner.get_default_kubeflow_metadata_config
  • 原experimental.KubeflowDagRunnerConfig替换为kubeflow_dag_runner.KubeflowDagRunnerConfig
  • 原experimental.KubeflowDagRunner替换为kubeflow_dag_runner.KubeflowDagRunner
• 补充原代码中缺失的logging导入和变量定义，避免二次报错

修复后完整代码 #

import logging
from tfx import v1 as tfx
import my_config
import my_pipeline1
from tfx.orchestration.kubeflow import kubeflow_dag_runner

# 补充缺失的变量定义，可根据实际路径修改
PIPELINE_ROOT = my_config.PIPELINE_ROOT
DATA_PATH = my_config.DATA_PATH

def run():
  """Define a kubeflow pipeline."""

  metadata_config = kubeflow_dag_runner.get_default_kubeflow_metadata_config()

  runner_config = kubeflow_dag_runner.KubeflowDagRunnerConfig(
      kubeflow_metadata_config=metadata_config,
      tfx_image=my_config.PIPELINE_IMAGE)

  kubeflow_dag_runner.KubeflowDagRunner(
      config=runner_config
  ).run(
      my_pipeline1.create_pipeline(
          pipeline_name=my_config.PIPELINE_NAME,
          pipeline_root=PIPELINE_ROOT,
          data_root=DATA_PATH,
          preprocessing_fn = my_config.PREPROCESSING_FN, 
          run_fn = my_config.RUN_FN,
      ))
if __name__ == '__main__':
  logging.set_verbosity(logging.INFO)
  run()

内容的提问来源于stack exchange，提问作者AshwinSrinivas_1