Back to Paradedb

Aggregate Syntax

docs/documentation/aggregates/overview.mdx

0.23.36.5 KB
Original Source

The pdb.agg function accepts an Elasticsearch-compatible JSON aggregate query string. It executes the aggregate using the columnar portion of the ParadeDB index, which can significantly accelerate performance compared to vanilla Postgres.

For example, the following query counts the total number of results for a search query.

<CodeGroup> ```sql SQL SELECT pdb.agg('{"value_count": {"field": "id"}}') FROM mock_items WHERE category === 'electronics'; ```
python
from paradedb import Agg, ParadeDB, Term

MockItem.objects.filter(
    category=ParadeDB(Term('electronics'))
).aggregate(agg=Agg('{"value_count": {"field": "id"}}'))
python
from sqlalchemy import select
from sqlalchemy.orm import Session
from paradedb.sqlalchemy import facets, pdb, search

stmt = (
    select(pdb.agg(facets.value_count(field="id")))
    .select_from(MockItem)
    .where(search.term(MockItem.category, "electronics"))
)

with Session(engine) as session:
    session.execute(stmt).all()
ruby
MockItem.search(:category)
        .term("electronics")
        .facets_agg(agg: ParadeDB::Aggregations.value_count(:id))
</CodeGroup>
ini
      agg
----------------
 {"value": 5.0}
(1 row)

This query counts the number of results for every distinct group:

<CodeGroup> ```sql SQL SELECT rating, pdb.agg('{"value_count": {"field": "id"}}') FROM mock_items WHERE category === 'electronics' GROUP BY rating ORDER BY rating LIMIT 5; ```
python
from paradedb import Agg, ParadeDB, Term

MockItem.objects.filter(
    category=ParadeDB(Term('electronics'))
).values('rating').annotate(
    agg=Agg('{"value_count": {"field": "id"}}')
).order_by('rating')[:5]
python
from sqlalchemy import select
from sqlalchemy.orm import Session
from paradedb.sqlalchemy import facets, pdb, search

stmt = (
    select(MockItem.rating, pdb.agg(facets.value_count(field="id")).label("agg"))
    .where(search.term(MockItem.category, "electronics"))
    .group_by(MockItem.rating)
    .order_by(MockItem.rating)
    .limit(5)
)

with Session(engine) as session:
    session.execute(stmt).all()
ruby
MockItem.search(:category)
        .term("electronics")
        .aggregate_by(
          :rating,
          agg: ParadeDB::Aggregations.value_count(:id)
        )
        .order(:rating)
        .limit(5)
</CodeGroup>
ini
 rating |      agg
--------+----------------
      3 | {"value": 1.0}
      4 | {"value": 3.0}
      5 | {"value": 1.0}
(3 rows)

Multiple Aggregations

To compute multiple aggregations at once, simply include multiple pdb.agg functions in the target list:

<CodeGroup> ```sql SQL SELECT pdb.agg('{"avg": {"field": "rating"}}') AS avg_rating, pdb.agg('{"value_count": {"field": "id"}}') AS count FROM mock_items WHERE category === 'electronics'; ```
python
from paradedb import Agg, ParadeDB, Term

MockItem.objects.filter(
    category=ParadeDB(Term('electronics'))
).aggregate(
    avg_rating=Agg('{"avg": {"field": "rating"}}'),
    count=Agg('{"value_count": {"field": "id"}}'),
)
python
from sqlalchemy import select
from sqlalchemy.orm import Session
from paradedb.sqlalchemy import facets, pdb, search

stmt = (
    select(
        pdb.agg(facets.avg(field="rating")).label("avg_rating"),
        pdb.agg(facets.value_count(field="id")).label("count"),
    )
    .select_from(MockItem)
    .where(search.term(MockItem.category, "electronics"))
)

with Session(engine) as session:
    session.execute(stmt).all()
ruby
MockItem.search(:category)
        .term("electronics")
        .facets_agg(
          avg_rating: ParadeDB::Aggregations.avg(:rating),
          count: ParadeDB::Aggregations.value_count(:id)
        )
</CodeGroup>
ini
   avg_rating   |     count
----------------+----------------
 {"value": 4.0} | {"value": 5.0}
(1 row)

Performance Optimization

On every query, ParadeDB runs checks to ensure that deleted or updated-away rows are not factored into the result set.

If your table is not frequently updated or you can tolerate an approximate result, the performance of aggregate queries can be improved by disabling these visibility checks. To do so, set the second argument of pdb.agg to false.

<CodeGroup> ```sql SQL SELECT pdb.agg('{"value_count": {"field": "id"}}', false) FROM mock_items WHERE description ||| 'running shoes'; ```
python
from paradedb import Agg, Match, ParadeDB

MockItem.objects.filter(
    description=ParadeDB(Match('running shoes', operator='OR'))
).aggregate(
    agg=Agg('{"value_count": {"field": "id"}}', exact=False)
)
python
from sqlalchemy import select
from sqlalchemy.orm import Session
from paradedb.sqlalchemy import facets, pdb, search

stmt = (
    select(pdb.agg(facets.value_count(field="id"), approximate=True).label("agg"))
    .where(search.match_any(MockItem.description, "running shoes"))
)

with Session(engine) as session:
    session.execute(stmt).all()
ruby
MockItem.search(:description)
        .matching_any("running shoes")
        .facets_agg(exact: false, agg: ParadeDB::Aggregations.value_count(:id))
</CodeGroup>

Disabling this check can improve query times by 2-4x in some cases (at the expense of correctness).

<Note> If a single query contains multiple `pdb.agg` calls, all of them must use the same visibility setting (either all `true` or all `false`). </Note>

JSON Fields

If metadata is a JSON field with key color, use metadata.color as the field name:

<CodeGroup> ```sql SQL SELECT pdb.agg('{"terms": {"field": "metadata.color"}}') FROM mock_items WHERE id @@@ pdb.all(); ```
python
from paradedb import Agg, All, ParadeDB

MockItem.objects.filter(
    id=ParadeDB(All())
).aggregate(agg=Agg('{"terms": {"field": "metadata.color"}}'))
python
from sqlalchemy import select
from sqlalchemy.orm import Session
from paradedb.sqlalchemy import facets, pdb, search

stmt = (
    select(pdb.agg(facets.terms(field="metadata.color")))
    .select_from(MockItem)
    .where(search.all(MockItem.id))
)

with Session(engine) as session:
    session.execute(stmt).all()
ruby
MockItem.search(:id)
        .match_all
        .facets_agg(agg: ParadeDB::Aggregations.terms("metadata.color"))
</CodeGroup> <Note> If a text or JSON field is used inside `pdb.agg`, it must use the [literal](/documentation/tokenizers/available-tokenizers/literal) or [literal normalized](/documentation/tokenizers/available-tokenizers/literal-normalized) tokenizer. </Note>