What is a working way to set up a custom Environment in AzureML using the Python SDK

Question

Below are my conda yml, Dockerfile and register.py(python SDK script) to create and register a custom environment in my AzureML Workspace. The AzureML documentation recommends this, if a custom system package should be installed. For me this is Graphviz. The conda environment and the docker file run successfully locally. However with my register.py script it fails:

66005-register.txt

66042-conda.txt

66033-dockerfile.txt

Answer 1

@Spranger Jordi Thanks for the question. Environments provide a way to manage software dependency so that controlled environments are reproducible with minimal manual configuration as you move between local and distributed cloud development environments. For more information about using environments for training and deployment with Azure Machine Learning, see Create and manage reusable environments.
Here is doc to add packages to an environment, Notebook to Train a model locally: Directly on your machine and within a Docker container.

AML currently sends a deserialized version of the Conda spec to Azure Container Registry for image building. This means that local artifacts (e.g., local pip packages, local requirements files) don't exist at image build time. There is a long-term approach to fixing this, anticipated to land in near future.