Use Bring Your Own Model (BYOM) to bring machine learning models, built using popular tools or platforms, and score them at scale in Vantage with no data movement. Supported formats are:
- PMML
- H2O (Open Source and Driverless AI)
- ONNX
- ONNXEmbeddings
- ONNXSeq2Seq
- ONNXClassification
- Dataiku
- DataRobot
- RealTime
- MLeap
BYOM is pre-installed on VantageCloud Lake and has the ability to score and predict on batch or new data on demand with the flexibility to scale hundreds of millions of records.
See Teradata Vantage™ - Bring Your Own Model User Guide.
Query execution location
By default, when the model table is in the Native Object Store (NOS) or the Object File System (OFS), the BYOM predict query runs on the compute cluster independent of the location of the data input table. The exceptions to this are the following:
- If there is no compute cluster configured, the query runs on the primary cluster.
- If the model table is stored in block storage, the BYOM predict query runs on the primary cluster.
- The BYOM predict query is forced to run in a specific compute group by configuring their user sessions to that specific compute group. See User Sessions for more information.