ParadeDB Docs

This is a beta feature available in versions 0.22.0 and above.

Creating a Composite Type

First, define a composite type whose field names and types match the columns you want to index:

CREATE TYPE item_fields AS (description TEXT, category TEXT, rating INTEGER);

Then reference the columns in a ROW() expression cast to the composite type:

CREATE INDEX search_idx ON mock_items
USING bm25 (id, (ROW(description, category, rating)::item_fields))
WITH (key_field='id');

Each field in the composite type is indexed as if it were a standalone column. Queries use the field names directly with the standard operators:

SELECT description, category FROM mock_items
WHERE description &&& 'running shoes';

Configuring Tokenizers

Fields in the composite type can use tokenizers and token filters by specifying them as the field type:

CREATE TYPE item_fields AS (
    description pdb.simple('stemmer=english'),
    category pdb.literal,
    in_stock BOOLEAN
);

CREATE INDEX search_idx ON mock_items
USING bm25 (id, (ROW(description, category, in_stock)::item_fields))
WITH (key_field='id');

Constraints

The following are not supported and will produce an error:

Anonymous ROW expressions: ROW(a, b) without a type cast is not allowed. Always cast to a named composite type.

Nested composites: A composite type cannot contain another composite type as a field.

Duplicate field names: Field names must be unique within a single BM25 index, across all composite types and regular columns in that index.

Documentation

Indexing 32+ Columns

Creating a Composite Type

Configuring Tokenizers

Constraints

​Creating a Composite Type

​Configuring Tokenizers

​Constraints

Creating a Composite Type

Configuring Tokenizers

Constraints