Teradata Vantage™ is more than a high-performance, scalable data platform, it is also an open analytics platform, offering a library of built-in analytic functions and extensible frameworks to enable our customers to integrate with their full ecosystem of analytics tools and platforms.
Teradata integrations include client packages such as Teradata Package for R (tdplyr) and Teradata Package for Python (teradataml) for R and Python developers to connect and access data and analytics functions within Vantage, Bring your own Model (BYOM) for executing third-party analytic models in-database, Script Table Operator (STO) for executing R and Python scripts in-database for data transformation and analytics, and now an API integration function to connect Vantage data with external analytic platforms, using their web services API interface, all initiated with a Vantage query.
The API is composed of two separate components.
The first component is an in-database function (a Table Operator named API_Request) that enables Vantage users to predict (score) machine learning models that are built on third party partner platforms against Vantage data through a Vantage query. For example, the 1.4.0.0 release supports Amazon SageMaker, Azure Machine Learning, Azure OpenAI, Google Vertex AI, and OpenAI model endpoints, with data passed from Vantage to these external platforms using a web services call, and results returned to the Vantage end user as results from the query.
The second component is a Python library (named tdapiclient), which is a companion package of the Teradata Package for Python (teradataml), Teradata's Python package for client-side processing. The tdapiclient Python packages allows Amazon SageMaker, Azure Machine Learning, and Google Vertex AI users of Vantage to call each CSP (cloud service provider)’s Python library interfaces to train and predict using data in Teradata Vantage tables. tdapiclient also transparently converts and copies the Teradata DataFrame to S3, Azure Blob Storage, or Google Cloud Storage for training as part of the fit() method which can invoke any of the API’s specified in the appropriate sections. For inference, it will use the data directly from Vantage input tables or queries via API_Request. Test inferences on small amounts of data can also be done directly through the client library. Finally, the tdapiclient also wraps the required BYOM calls to productionize the scoring process.
