Azure Machine Learning is a cloud machine learning platform which offers services to train machine learning models using Azure compute resources and then host those models in Azure containers. These hosted models are called 'endpoint' and can be accessed using web address. They are divided into two categories:
- Online endpoint : Used for real-time scoring.
- Batch endpoint: Used for batch scoring.
The integration with Azure Machine Learning enables customers to do real-time scoring using Teradata tables or queries and online endpoint.
Models created using Azure Machine Learning can be downloaded locally. Teradata customers have the option to either use Azure Machine Learning for online scoring, or insert these models in Vantage and do the scoring using the BYOM feature.
The integration with Azure Machine Learning includes the following:
- API_Request In-database Function
This in-database function enables Vantage users to connect with an Azure Machine Learning endpoint for scoring using a Vantage query (SQL).
- Teradata Package for Python (teradataml) Azure Machine Learning Extension Library (tdapiclient)
This client package allows data scientists and developers to use their Python development environment to connect to Vantage and use easier API to train and predict using teradataml Dataframe objects with Teradata tables or queries.
API_Request In-database Function
Typical steps to use API_Request in-database function:
- Data cleaning, feature engineering inside the database;
- Exporting data for training and model fitting;
- Azure Machine Learning model training and model deployment;
- Collecting information about endpoint (data types, number of columns, authentication keys);
- Real-time scoring using Teradata API_Request in-database function with API_TYPE as "azure-ml".
teradataml Azure Extension Library (tdapiclient)
The teradataml Azure Extension Library (tdapiclient) uses Vantage DataFrame to train Azure Machine Learning models by transparently converting Vantage DataFrame into Azure Machine Learning Blob address
- Enable users to prepare the training data by leveraging Teradata’s Python based in-DB analytics functions.
- Leveraging teradataml and Azure Machine Learning Python SDK capabilities which allow users to create model over Azure Machine Learning directly from Vantage.
- Once created models are deployed as endpoint in Azure or over Vantage using ONNX inter-exchange format, business users can get access of Azure analytic services for real-time scoring directly from Vantage.