Prepare

initialize conda

conda init
exec bash

create and activate environment

conda env create -f environment.yml
conda activate mlops-workshop-env

set ipython kernel

ipython kernel install --user --name=mlops-workshop-env

login

export $(cat .env| grep -v "#" | xargs)
az login --use-device-code
az configure --defaults group=$RESOURCE_GROUP workspace=$AML_WORKSPACE_NAME
az configure -l -o table

pre-commit

pre-commit install

Step by Step

Step 0 : Run on AML

Execute base Jupyter notebook on Azure Machine Learning.

Step 1 : Use MLflow

Add MLflow experiment management.

Step 2 : Convert ipynb to py

Convert base Jupyter notebook to Python script.

jupyter nbconvert --to script "01_mlflow/model_build.ipynb"
mv 01_mlflow/model_build.py 02_python_script/build_model.py

Refactor and fix generated python script file, and run.

cd 02_python_script
python build_model.py
pytest test.py

(Recommend) For Step 3, add input_train_data and input_valid_data properties that mean csv file path.

Step 3 : Create AML Job

Define dependency-assets and execute Python script on Azure Machine Learning as a Job.

az ml data create -f ./03_job/data-train.yml
az ml data create -f ./03_job/data-valid.yml
az ml job create -f ./03_job/job-train.yml

Step 4 : Create AML Pipeline

Define pipeline and reproduce training job.

az ml job create -f 04_pipeline/pipeline.yml

Step 5 : Deploy model as API

az ml environment create -f environment.yml
az ml online-endpoint create -f online-endpoint.yml
az ml online-deployment create -n version-01 -f ./05_deploy/online-deployment.yml

update environment

conda env update -f environment.yml