search

microsoft / solution-accelerator-many-models

MIT License
193 stars 85 forks source link

Error When Running Forecast Pipeline #139

Closed gantonius closed 2 years ago

gantonius commented 2 years ago

I am currently following this notebook and now encountering an error below when the inference experiment is run on AML:

Response status code does not indicate success: 400 (BaseImage, BaseDockerfile, or BuildContext must be set for Docker-based environments.). Microsoft.RelInfra.Common.Exceptions.ErrorResponseException: BaseImage, BaseDockerfile, or BuildContext must be set for Docker-based environments.

It failed in this step of the AzureML pipeline: image

The following are the versions of the AML libraries I used:

Is this perhaps related to this PR where the environment needs to be explicitly specified in the training and inference settings?

Can you help to diagnose this issue? Thanks.

gantonius commented 2 years ago

It seems that's right that I will need to provide inference_env=Environment.get(ws, name='AzureML-AutoML') in the AutoMLPipelineBuilder.get_many_models_batch_inference_steps parameter as indicated by this PR.

I didn't provide the train_env when calling AutoMLPipelineBuilder.get_many_models_train_steps but provided the inference_env' inAutoMLPipelineBuilder.get_many_models_batch_inference_steps` and the training and inference runs are successful.