Configure generators
Set table relationships

Set table relationships

For multi-table datasets, you can set the relationships between tables in your generator. The generator can then learn to retain the correlations and referential integrity of your original dataset.

The basic scenario is the two-table relationship, while the more complex but also easy to set up are multi-table relationships.


MOSTLY AI differentiates between two types of table relationships: context and non-context relationships. The relationships are based on the primary and foreign key concepts and define how MOSTLY AI trains generators to retain the correlations and referential integrity in multi-table datasets.

Primary key

The goal of a primary key is to uniquely identify a record. In database schema design, a primary key constraint sets a requirement in a column that the column must contain a unique value for each record.

Foreign key

A foreign key establishes a relationship between two tables. In database design, the foreign key constraint requires that each record from a table is related to a record from another (usually by pointing to the other table's primary key).

Context foreign key

In MOSTLY AI, you can set a context foreign key to define that a table is trained on and then generated in the context of another table. With a context foreign key, you can fully retain the correlations and maintain the referential integrity between two tables and across a hierarchy of grandparent - parent - child tables.

A table can have only one context foreign key. If the same table has multiple foreign keys in your original data, you can use regular foreign keys for the remaining relationships.

Non-context foreign key

In MOSTLY AI, a non-context foreign key helps to maintain the referential integrity between two tables. The correlations between such tables are retained on a "best effort" basis.

Regular foreign keys are useful in the following scenarios.

  • Multiple foreign keys in a single table. If a table has multiple foreign keys, you can prioritize one as the context (to retain correlations and referential integrity), and have the remaining as regular foreign keys (to retain the referential integrity).
  • Relationship between two subject tables. If you have two subject tables, you can retain their referential integrity with a non-context foreign key.