Open
Description
When creating e.g. the ACI deploy, we pass the parameters inferencePath which points to $(Build.SourcesDirectory)/$(SOURCES_DIR_TRAIN)/scoring/inference_config.yml. This file contains parameters for the inference config. On the other hand, if one enables the "deploy to azure web app", then the docker image is made by the file ml_service/util/create_scoring_image.py, which does not read (or gets passed variables from) inference_config.yml. It would be better if one could define these variable only once (in inference_config.yml). In our case we needed to customize the "sources directory", which at the moment forces us to change the script create_scoring_image.py