add tests edit bounds

Author: rbs333

redisvl/extensions/llmcache/semantic.py

@@ -238,9 +238,9 @@ def set_threshold(self, distance_threshold: float) -> None:
         Raises:
             ValueError: If the threshold is not between 0 and 1.
         """
-        if not 0 <= float(distance_threshold) <= 1:
+        if not 0 <= float(distance_threshold) <= 2:
             raise ValueError(
-                f"Distance must be between 0 and 1, got {distance_threshold}"
+                f"Distance must be between 0 and 2, got {distance_threshold}"
             )
         self._distance_threshold = float(distance_threshold)
 

redisvl/index/index.py

@@ -16,12 +16,7 @@
     Union,
 )
 
-from redisvl.utils.utils import (
-    deprecated_argument,
-    deprecated_function,
-    norm_cosine_distance,
-    sync_wrapper,
-)
+from redisvl.utils.utils import deprecated_argument, deprecated_function, sync_wrapper
 
 if TYPE_CHECKING:
     from redis.commands.search.aggregation import AggregateResult
@@ -35,13 +30,7 @@
 
 from redisvl.exceptions import RedisModuleVersionError, RedisSearchError
 from redisvl.index.storage import BaseStorage, HashStorage, JsonStorage
-from redisvl.query import (
-    BaseQuery,
-    CountQuery,
-    FilterQuery,
-    VectorQuery,
-    VectorRangeQuery,
-)
+from redisvl.query import BaseQuery, BaseVectorQuery, CountQuery, FilterQuery
 from redisvl.query.filter import FilterExpression
 from redisvl.redis.connection import (
     RedisConnectionFactory,
@@ -92,9 +81,7 @@ def process_results(
         and not query._return_fields  # type: ignore
     )
 
-    if (
-        isinstance(query, VectorQuery) or isinstance(query, VectorRangeQuery)
-    ) and query._normalize_vector_distance:
+    if (isinstance(query, BaseVectorQuery)) and query._normalize_vector_distance:
         dist_metric = VectorDistanceMetric(
             schema.fields[query._vector_field_name].attrs.distance_metric.upper()  # type: ignore
         )

redisvl/query/__init__.py

@@ -1,5 +1,6 @@
 from redisvl.query.query import (
     BaseQuery,
+    BaseVectorQuery,
     CountQuery,
     FilterQuery,
     RangeQuery,
@@ -9,6 +10,7 @@
 
 __all__ = [
     "BaseQuery",
+    "BaseVectorQuery",
     "VectorQuery",
     "FilterQuery",
     "RangeQuery",

redisvl/query/query.py

@@ -5,6 +5,7 @@
 
 from redisvl.query.filter import FilterExpression
 from redisvl.redis.utils import array_to_buffer
+from redisvl.utils.utils import denorm_cosine_distance
 
 
 class BaseQuery(RedisQuery):
@@ -175,6 +176,8 @@ class BaseVectorQuery:
     DISTANCE_ID: str = "vector_distance"
     VECTOR_PARAM: str = "vector"
 
+    _normalize_vector_distance: bool = False
+
 
 class HybridPolicy(str, Enum):
     """Enum for valid hybrid policy options in vector queries."""
@@ -516,6 +519,14 @@ def set_distance_threshold(self, distance_threshold: float):
             raise TypeError("distance_threshold must be of type float or int")
         if distance_threshold < 0:
             raise ValueError("distance_threshold must be non-negative")
+        if self._normalize_vector_distance:
+            if distance_threshold > 1:
+                raise ValueError(
+                    "distance_threshold must be between 0 and 1 when normalize_vector_distance is set to True"
+                )
+
+            # User sets normalized value 0-1 denormalize for use in DB
+            distance_threshold = denorm_cosine_distance(distance_threshold)
         self._distance_threshold = distance_threshold
 
         # Reset the query string

redisvl/utils/utils.py

@@ -197,7 +197,12 @@ def norm_cosine_distance(value: float) -> float:
     """
     Normalize the cosine distance to a similarity score between 0 and 1.
     """
-    return (2 - value) / 2
+    return max((2 - value) / 2, 0)
+
+
+def denorm_cosine_distance(value: float) -> float:
+    """Denormalize the distance threshold from [0, 1] to [0, 1] for our db."""
+    return max(2 - 2 * value, 0)
 
 
 def norm_l2_distance(value: float) -> float:

tests/integration/test_query.py

@@ -4,7 +4,7 @@
 from redis.commands.search.result import Result
 
 from redisvl.index import SearchIndex
-from redisvl.query import CountQuery, FilterQuery, RangeQuery, VectorQuery
+from redisvl.query import CountQuery, FilterQuery, VectorQuery, VectorRangeQuery
 from redisvl.query.filter import (
     FilterExpression,
     Geo,
@@ -105,7 +105,7 @@ def sorted_filter_query():
 
 @pytest.fixture
 def normalized_range_query():
-    return RangeQuery(
+    return VectorRangeQuery(
         vector=[0.1, 0.1, 0.5],
         vector_field_name="user_embedding",
         normalize_vector_distance=True,
@@ -117,7 +117,7 @@ def normalized_range_query():
 
 @pytest.fixture
 def range_query():
-    return RangeQuery(
+    return VectorRangeQuery(
         vector=[0.1, 0.1, 0.5],
         vector_field_name="user_embedding",
         return_fields=["user", "credit_score", "age", "job", "location"],
@@ -127,7 +127,7 @@ def range_query():
 
 @pytest.fixture
 def sorted_range_query():
-    return RangeQuery(
+    return VectorRangeQuery(
         vector=[0.1, 0.1, 0.5],
         vector_field_name="user_embedding",
         return_fields=["user", "credit_score", "age", "job", "location"],
@@ -272,7 +272,7 @@ def test_search_and_query(index):
 
 
 def test_range_query(index):
-    r = RangeQuery(
+    r = VectorRangeQuery(
         vector=[0.1, 0.1, 0.5],
         vector_field_name="user_embedding",
         return_fields=["user", "credit_score", "age", "job"],
@@ -343,7 +343,7 @@ def search(
             assert doc.location == location
 
     # if range query, test results by distance threshold
-    if isinstance(query, RangeQuery):
+    if isinstance(query, VectorRangeQuery):
         for doc in results.docs:
             print(doc.vector_distance)
             assert float(doc.vector_distance) <= distance_threshold
@@ -354,7 +354,7 @@ def search(
 
     # check results are in sorted order
     if sort:
-        if isinstance(query, RangeQuery):
+        if isinstance(query, VectorRangeQuery):
             assert [int(doc.age) for doc in results.docs] == [12, 14, 18, 100]
         else:
             assert [int(doc.age) for doc in results.docs] == [
@@ -370,7 +370,7 @@ def search(
 
 @pytest.fixture(
     params=["vector_query", "filter_query", "range_query"],
-    ids=["VectorQuery", "FilterQuery", "RangeQuery"],
+    ids=["VectorQuery", "FilterQuery", "VectorRangeQuery"],
 )
 def query(request):
     return request.getfixturevalue(request.param)
@@ -778,3 +778,15 @@ def test_range_query_normalize_cosine_distance(index, normalized_range_query):
 
     for r in res:
         assert 0 <= float(r["vector_distance"]) <= 1
+
+
+def test_range_query_normalize_bad_input(index):
+    with pytest.raises(ValueError):
+        VectorRangeQuery(
+            vector=[0.1, 0.1, 0.5],
+            vector_field_name="user_embedding",
+            normalize_vector_distance=True,
+            return_score=True,
+            return_fields=["user", "credit_score", "age", "job", "location"],
+            distance_threshold=1.2,
+        )

tests/unit/test_utils.py

@@ -16,6 +16,7 @@
 )
 from redisvl.utils.utils import (
     assert_no_warnings,
+    denorm_cosine_distance,
     deprecated_argument,
     deprecated_function,
     norm_cosine_distance,
@@ -28,6 +29,17 @@ def test_norm_cosine_distance():
     assert norm_cosine_distance(input) == expected
 
 
+def test_denorm_cosine_distance():
+    input = 0
+    expected = 2
+    assert denorm_cosine_distance(input) == expected
+
+
+def test_norm_denorm_cosine():
+    input = 0.6
+    assert input == round(denorm_cosine_distance(norm_cosine_distance(input)), 6)
+
+
 def test_even_number_of_elements():
     """Test with an even number of elements"""
     values = ["key1", "value1", "key2", "value2"]