Azure AutoML

Defining Azure Automated Machine Learning

• Azure offers a serviced called AutoML for automated machine learning
• AutoML automates hyperparameter tuning and model selection

• It achieves this by:

  1. AzureML runs all of our pickle models concurrently
  2. These results are compared
  3. The best model is recommended for the job based on the training scores

Describing the Benefits of AutoML

• AutoML doesn't require models to be already deployed

  • They just can use AutoML in the model training process
• The AutoML HyperDrive service finds optimal hyperparameter values automatically

• Azure Databricks has its own AutoML capabilities

  • Specifically, it uses MLFlow
  • Thus, it doesn't really need AutoML from AzureML

Steps for using AutoML in Azure

1. Select your experiment type

   • This could be classification, regression, or forecasting

2. Ingest the data

   • Data can be ingested from:

     • Our local computer
     • Cloud storage
   • A popular storage service is Azure Blob Storage

3. Configure the targets

   • The compute targets must be configured for running the experiment

   • This can be a local machine or a cloud resource:

     • Azure Machine Learning Computer
     • Azure HDInsight
     • A remote virtual machine

4. Configure the AutoML job

   • The parameters must be configured as AzureML iterates over different models and hyperparameter settings
   • We also identify which metrics AutoML should look at to determine the best model
5. Submit the training run

References

• Azure Data Science Certification
• Defining MLOps with Python
• Describing Model Registrey and Deployment
• Describing Pipelines in Azure
• Describing Azure HDInsight
• Comparing Azure HDInsight and Databricks