Data Lab
Data Lab connects your source data to the platform and prepares it for modelling. The flow has two main stages: ingestion reads and validates source files to create a datasource; onboarding prepares that datasource to create a training dataset.
A datasource records imported data. A dataset adds the choices needed for a prediction task, including the target, selected features and preprocessing. Completing ingestion makes data available for onboarding; it does not train a model.
How data moves through the platform
Prepare source files
Organise consistent CSV or Parquet files and decide which columns are features, targets or ignored values. Start with Prepare data.
Connect source storage
A storage connector gives the platform access to the S3 bucket containing your files.
Define the schema
A data schema describes the source columns, their types and their roles in modelling. Review the retrieved fields before saving it.
Ingest the data
An ingestion job applies a configuration to files in storage and creates or updates a datasource.
Onboard the dataset
An onboarding job applies the prediction and preprocessing choices to the datasource and creates or updates a dataset.
Storage and compute
Source storage holds the files you supply. Account storage holds resources created by the platform. These roles have different permission requirements: source-data access normally uses a read-only storage connector. Dedicated buckets are recommended even when the same bucket could serve both roles.
Ingestion and onboarding jobs use a processing compute configuration. The configuration specifies the machine resources; the job specifies which data to process and how long it may run.
Feature Transform Services also appear in Data Lab. They support prediction-time transformations for models that use feature engineering. The deployment workflow explains when to select one.
Inspect the data before training
Use lineage to trace a datasource back to its ingestion history or a dataset back to its datasource and onboarding job. Dataset distributions show how values are represented, including categorical frequencies, numeric bins and target statistics. These checks help confirm that the data matches the task before you train a model.
In the app, ingestion resources are grouped under Data Lab > Datasources and onboarding resources under Data Lab > Datasets. See Datasources and Inspect datasets for the inspection workflows.
Next steps
Start with Prepare data, then connect storage and define the schema.