Adds l1 distance, added on 0.7.0

.github/workflows/tests.yml

@@ -8,7 +8,7 @@ jobs:
     strategy:
         matrix:
             python-version: ['3.8', '3.9', '3.10', '3.11']
-            postgres-version: ['15.1.0.118']
+            postgres-version: ['15.1.1.78']
 
     services:
 

docs/api.md

@@ -89,6 +89,7 @@ Available options for distance `measure` are:
 
 - `vecs.IndexMeasure.cosine_distance`
 - `vecs.IndexMeasure.l2_distance`
+- `vecs.IndexMeasure.l1_distance`
 - `vecs.IndexMeasure.max_inner_product`
 
 which correspond to different methods for comparing query vectors to the vectors in the database.

docs/concepts_indexes.md

@@ -26,6 +26,7 @@ Available options for distance `measure` are:
 
 - `vecs.IndexMeasure.cosine_distance`
 - `vecs.IndexMeasure.l2_distance`
+- `vecs.IndexMeasure.l1_distance`
 - `vecs.IndexMeasure.max_inner_product`
 
 which correspond to different methods for comparing query vectors to the vectors in the database.

docs/index.md

@@ -30,7 +30,7 @@ Vecs is a Python client library for managing and querying vector stores in Postg
 ## Overview
 
 - Vector Management: create collections to persist and update vectors in a PostgreSQL database.
-- Querying: Query vectors efficiently using measures such as cosine distance, l2 distance, or max inner product.
+- Querying: Query vectors efficiently using measures such as cosine distance, l2 distance, l1 distance, or max inner product.
 - Metadata: Each vector can have associated metadata, which can also be used as filters during queries.
 - Hybrid Data: vecs creates its own schema and can coexist with your existing relational data
 

setup.py

@@ -40,7 +40,7 @@ def read_package_variable(key, filename="__init__.py"):
 long_description = (Path(__file__).parent / "README.md").read_text()
 
 REQUIRES = [
-    "pgvector==0.1.*",
+    "pgvector==0.3.*",
     "sqlalchemy==2.*",
     "psycopg2-binary==2.9.*",
     "flupy==1.*",

src/tests/conftest.py

@@ -21,7 +21,7 @@ def maybe_start_pg() -> Generator[None, None, None]:
     to using the PYTEST_DB connection string"""
 
     container_name = "vecs_pg"
-    image = "supabase/postgres:15.1.0.118"
+    image = "supabase/postgres:15.1.1.78"
 
     connection_template = "postgresql://{user}:{pw}@{host}:{port:d}/{db}"
     conn_args = parse(connection_template, PYTEST_DB)

src/tests/test_collection.py

@@ -806,6 +806,19 @@ def test_l2_index_query(client: vecs.Client) -> None:
     assert len(results) == 1
 
 
+def test_l1_index_query(client: vecs.Client) -> None:
+    dim = 4
+    bar = client.get_or_create_collection(name="bar", dimension=dim)
+    bar.upsert([("a", [1, 2, 3, 4], {})])
+    bar.create_index(measure=vecs.IndexMeasure.l1_distance)
+    results = bar.query(
+        data=[1, 2, 3, 4],
+        limit=1,
+        measure="l1_distance",
+    )
+    assert len(results) == 1
+
+
 def test_max_inner_product_index_query(client: vecs.Client) -> None:
     dim = 4
     bar = client.get_or_create_collection(name="bar", dimension=dim)

src/vecs/__init__.py

@@ -9,7 +9,7 @@
 )
 
 __project__ = "vecs"
-__version__ = "0.4.4"
+__version__ = "0.4.5"
 
 
 __all__ = [

src/vecs/collection.py

@@ -82,6 +82,7 @@ class IndexMeasure(str, Enum):
     cosine_distance = "cosine_distance"
     l2_distance = "l2_distance"
     max_inner_product = "max_inner_product"
+    l1_distance = "l1_distance"
 
 
 @dataclass
@@ -124,12 +125,14 @@ class IndexArgsHNSW:
     IndexMeasure.cosine_distance: "vector_cosine_ops",
     IndexMeasure.l2_distance: "vector_l2_ops",
     IndexMeasure.max_inner_product: "vector_ip_ops",
+    IndexMeasure.l1_distance: "vector_l1_ops",
 }
 
 INDEX_MEASURE_TO_SQLA_ACC = {
     IndexMeasure.cosine_distance: lambda x: x.cosine_distance,
     IndexMeasure.l2_distance: lambda x: x.l2_distance,
     IndexMeasure.max_inner_product: lambda x: x.max_inner_product,
+    IndexMeasure.l1_distance: lambda x: x.l1_distance,
 }