@satonaoki What is the reason for making this change? We provide both training scripts as examples, and users are welcome to try either. train.py is the simplest and can be run locally, so it been set as the default.

@j-so

(1) When running the training script locally without Azure ML pipeline, you can simply run train.py directly.

'TRAIN_SCRIPT_PATH' environment variable is not used in this scenario.

(2) When running diabetes_regression_build_train_pipeline.py locally to publish a new Azure ML pipeline as described here, 'TRAIN_SCRIPT_PATH' environment variable in .env is used.

So I think this should be train_aml.py, not train.py. train.py does not work with diabetes_regression_build_train_pipeline.py.

(3) When running diabetes_regression_build_train_pipeline.py from Azure DevOps pipeline, 'TRAIN_SCRIPT_PATH' environment variable in diabetes_regression-variables-template.yml is used. It is already set as 'train_aml.py'.

I see - looks like train.py is now out of date since it was introduced. I agree then that we should switch to train_aml for the local example environment variables.