Batch Inference
You'll score data with a registered model, using pandas for quick checks or Spark UDFs for production scale, in ~10 min.
Prereqs: Save a Model to Unity Catalog
The big picture
A batch job that loads a model from Unity Catalog by alias, scores rows, and writes predictions to a Delta table.
Prerequisites
- A model registered in Unity Catalog (see Save a Model to Unity Catalog).
- Data available in a Delta table or DataFrame to run inference against.
Inference methods
| Method | Best for | Reference |
|---|---|---|
| 🥇 Spark UDF | Production-scale tabular data distributed across a cluster | How to speed up inference |
| Pandas single-node | Small batches, quick validation, single-node notebooks | |
| Pandas UDF for images | Computer vision workloads with images stored in object storage | Distributed inference with pandas UDF |
Next
- Do next: Datasets as Feature Tables
- Learn why: Data science
- Reference: Model lifecycle in Unity Catalog, How to speed up inference, Image ETL and inference