Datasources

A datasource is the output of ingestion. Dataset onboarding uses it to create data prepared for training. Check the datasource before selecting it for an onboarding job, particularly when several imports or increments have similar names.

Before you begin

Complete an ingestion job and retain the output datasource name or ID.

Inspect a datasource

1. Open Data Lab > Datasources > Datasources and locate the output record.
2. Check that its status is AVAILABLE. Review Increment, Created at and Updated at in the list.
3. Open the datasource and inspect Overview. Check its ID, schema and processing references against the ingestion job.
4. Open Lineage to trace the datasource through its processing steps. Select a node to inspect its metadata.
5. Use Details to inspect the complete resource record when you need identifiers or increment information.

Use a datasource for onboarding

Select the intended datasource in an onboarding job, together with an onboarding configuration and processing compute configuration. Confirm that the configuration’s target and prediction type match the schema used for this data.

For incremental data, check both the datasource and dataset context. Ingesting additional data and onboarding it are separate jobs; completing ingestion alone does not update a training dataset or retrain a model.

Next steps

Prepare an onboarding configuration and run an onboarding job.