ml-repository-azureml-cli-v2

Sample repository for MLOps (especially Continuous Deployment or Training) with Azure Machine Learning.

Dependencies

• Python >= 3.8
• conda

Preparation

Create conda environment

conda env create -f environment.yml
conda activate env-ml-reposiotry-azureml-cli-v2

pre-commit

pre-commit install

Create resources to execute job

Set below variables for your environment.

• subscription_id = "subscription_id"
• resource_group = "resource_group_name"
• workspace_name = "ml_workspace_name"

python prepare.py

prepare.py script create a compute cluster, a dataset for regression and an environment optionally.

Job execution

az ml job create -f ./job/search_hyperparameter.yml -g <resource_group> -w <ml_workspace>

Continuous Training

Required

1. Register application and grant Azure ML access.
2. Set below variables from registered application and Azure ML workspace as Github secrets.
   1. AZURE_CLIENT_ID
   2. AZURE_TENANT_ID
   3. AZURE_SUBSCRIPTION_ID
   4. AZURE_RESOURCE_GROUP_NAME
   5. AZURE_ML_WORKSPACE_NAME

https://docs.microsoft.com/en-us/azure/developer/github/connect-from-azure?tabs=azure-portal%2Clinux

Code quality

To maintain code quality, below libraries are used in this repository.

• flake8 : lint based on PEP8
• black : auto-format based on PEP8 (flake8)
• isort : auto-sort import
• mypy : check type
• pre-commit : check code before commit

Code quality is important for team development.

Reference

https://docs.microsoft.com/en-us/azure/machine-learning/how-to-train-cli#sweep-hyperparameters https://docs.microsoft.com/en-us/azure/machine-learning/how-to-create-register-data-assets?tabs=Python-SDK#uris