Transforms
How to change, enrich, filter, and type episode rows in Refiner
Transforms are pipeline steps between readers and writers. They can operate on Python rows, async I/O, batches, or Arrow-backed vectorized tables.
| Transform style | Use it for |
|---|---|
| Row Transforms | Simple episode-level logic. |
| Async and Batch Transforms | API calls, model calls, batched perception. |
| Vectorized Transforms | Fast columnar filtering, projection, renaming, and computed columns. |
| Expressions | The expression DSL behind vectorized transforms. |
| Schemas and DTypes | Type hints for assets, media, and writer schemas. |
| GPU Video Transcoding | NVENC video conversion and ordered image sequence encoding. |
For packaged episode-level workflows, see Episode Operations.