Data domains¶

A data domain is a project-level namespace that isolates features and processes inside a shared feature store. Two teams can register a feature called total_amount for the same CUSTOMER_ID entity without colliding — the catalog stores DATA_DOMAIN as part of every key.

The three helpers¶

Always use the dedicated functions — never assign tdfs4ds.DATA_DOMAIN directly:

Function	Purpose	Side effect
`tdfs4ds.create_data_domain(name)`	Create and activate a new domain. Idempotent (warns if it already exists).	Sets `tdfs4ds.DATA_DOMAIN = name`.
`tdfs4ds.select_data_domain(name)`	Activate an existing domain.	Sets `tdfs4ds.DATA_DOMAIN`; raises `ValueError` if the domain is unknown.
`tdfs4ds.get_data_domains()`	List every domain (the active one is marked `*`).	None.

Setup vs analytics¶

The convention used in the bundled notebooks:

Notebook role	Function to use
Bootstrap / setup notebook	`create_data_domain` — first-time creation
Analytics / modelling notebook	`select_data_domain` — activate the existing one

select_data_domain raises rather than silently creating a typo'd domain.

Cross-domain features¶

Every public query (catalog reads, build_dataset, the consumer agent) is scoped to the active domain. To pull features from two domains in the same dataset, build two datasets and join the views manually — build_dataset does not accept multiple domains by design.

Storage layout¶

Every domain shares the same physical catalog tables (FS_FEATURE_CATALOG, FS_PROCESS_CATALOG, …) — the DATA_DOMAIN column inside each table is the isolation key.

tdfs4ds.DATA_DOMAIN = 'PROJECT_A'
tdfs4ds.feature_catalog().shape   # PROJECT_A only

tdfs4ds.select_data_domain('PROJECT_B')
tdfs4ds.feature_catalog().shape   # PROJECT_B only