Building Production Pipelines WIth AzureML and MLRun

• This demo works with the online feature store, which is currently not part of the Open Source default deployment.

In This demo we show how to:

0. Set up the environment
1. Use MLRun Feature Store to ingest and prepare data
2. Create offline feature vector (snapshot) for training
3. Run AzureML AutoML Service as an automated step (function) in MLRun:

• Take a snapshot of the offline feature vector and register as an AzureDL dataset
• Initialize the required resources, experiments and AutoML job in AzureML and track its progress
• Retrieve the generated models along with their results and register in MLRun

4. View and compare the AzureML Models using MLRun tools
5. Build Real-time Serving pipeline with multiple stages:

• Accept and parse requests
• Enrich and impute with real-time features (from the feature store)
• Predict using an ensemble of 3 models (Generated by AzureML)
• Respond with average prediction

6. Provide real-time model monitoring

Notebook

The demo has a single notebook:

• azureml_demo.ipynb — implement all the steps outlined above.