Field of Vector length

Signed-off-by: jyejare <jyejare@redhat.com>
feast-dev · franciscojavierarceo · Apr 23, 2025 · Apr 16, 2025 · Apr 16, 2025 · Apr 21, 2025
commit 93cb9f5744f47f98ae8839def8cb22076ab3e4eb
@@ -42,4 +42,7 @@ message FeatureSpecV2 {
 
     // Metric used for vector similarity search.
     string vector_search_metric = 6;
+
+    // Field indicating the vector length
+    int32 vector_len = 7;
 }
@@ -1555,6 +1555,18 @@ def _get_feature_view_and_df_for_online_write(
                 except Exception as _:
                     raise DataFrameSerializationError(df)
 
+        if feature_view.features[0].vector_index and df is not None:
+            fv_vector_feature_name = feature_view.features[0].name
+            df_vector_feature_index = df.columns.get_loc(fv_vector_feature_name)
+            if feature_view.features[0].vector_len != 0:
+                if (
+                    feature_view.features[0].vector_len
+                    != df.shape[df_vector_feature_index]
+                ):
+                    raise ValueError(
+                        f"The dataframe for {fv_vector_feature_name} column has {df.shape[1]} vectors, but the feature view {feature_view.name} expects {feature_view.features[0].vector_len}"
+                    )
+
         # # Apply transformations if this is an OnDemandFeatureView with write_to_online_store=True
         if (
             isinstance(feature_view, OnDemandFeatureView)

@@ -33,6 +33,7 @@ class Field:
         description: A human-readable description.
         tags: User-defined metadata in dictionary form.
         vector_index: If set to True the field will be indexed for vector similarity search.
+        vector_len: The length of the vector if the vector index is set to True.
         vector_search_metric: The metric used for vector similarity search.
     """
 
@@ -41,6 +42,7 @@ class Field:
     description: str
     tags: Dict[str, str]
     vector_index: bool
+    vector_len: int
     vector_search_metric: Optional[str]
 
     def __init__(
@@ -51,6 +53,7 @@ def __init__(
         description: str = "",
         tags: Optional[Dict[str, str]] = None,
         vector_index: bool = False,
+        vector_len: int = 0,
         vector_search_metric: Optional[str] = None,
     ):
         """
@@ -69,6 +72,7 @@ def __init__(
         self.description = description
         self.tags = tags or {}
         self.vector_index = vector_index
+        self.vector_len = vector_len
         self.vector_search_metric = vector_search_metric
 
     def __eq__(self, other):
@@ -80,6 +84,7 @@ def __eq__(self, other):
             or self.dtype != other.dtype
             or self.description != other.description
             or self.tags != other.tags
+            or self.vector_len != other.vector_len
             # or self.vector_index != other.vector_index
             # or self.vector_search_metric != other.vector_search_metric
         ):
@@ -100,6 +105,7 @@ def __repr__(self):
             f"    description={self.description!r},\n"
             f"    tags={self.tags!r}\n"
             f"    vector_index={self.vector_index!r}\n"
+            f"    vector_len={self.vector_len!r}\n"
             f"    vector_search_metric={self.vector_search_metric!r}\n"
             f")"
         )
@@ -117,6 +123,7 @@ def to_proto(self) -> FieldProto:
             description=self.description,
             tags=self.tags,
             vector_index=self.vector_index,
+            vector_len=self.vector_len,
             vector_search_metric=vector_search_metric,
         )
 
@@ -131,12 +138,14 @@ def from_proto(cls, field_proto: FieldProto):
         value_type = ValueType(field_proto.value_type)
         vector_search_metric = getattr(field_proto, "vector_search_metric", "")
         vector_index = getattr(field_proto, "vector_index", False)
+        vector_len = getattr(field_proto, "vector_len", 0)
         return cls(
             name=field_proto.name,
             dtype=from_value_type(value_type=value_type),
             tags=dict(field_proto.tags),
             description=field_proto.description,
             vector_index=vector_index,
+            vector_len=vector_len,
             vector_search_metric=vector_search_metric,
         )
 

@@ -311,7 +311,7 @@ def update(
             for table in tables_to_keep:
                 table_name = _table_id(project, table)
                 if config.online_store.vector_enabled:
-                    vector_value_type = f"vector"
+                    vector_value_type = "vector"
                 else:
                     # keep the vector_value_type as BYTEA if pgvector is not enabled, to maintain compatibility
                     vector_value_type = "BYTEA"

@@ -55,6 +55,7 @@ class FeatureSpecV2(google.protobuf.message.Message):
     DESCRIPTION_FIELD_NUMBER: builtins.int
     VECTOR_INDEX_FIELD_NUMBER: builtins.int
     VECTOR_SEARCH_METRIC_FIELD_NUMBER: builtins.int
+    VECTOR_LEN_FIELD_NUMBER: builtins.int
     name: builtins.str
     """Name of the feature. Not updatable."""
     value_type: feast.types.Value_pb2.ValueType.Enum.ValueType
@@ -68,6 +69,8 @@ class FeatureSpecV2(google.protobuf.message.Message):
     """Field indicating the vector will be indexed for vector similarity search"""
     vector_search_metric: builtins.str
     """Metric used for vector similarity search."""
+    vector_len: builtins.int
+    """Field indicating the vector length"""
     def __init__(
         self,
         *,
@@ -77,7 +80,8 @@ class FeatureSpecV2(google.protobuf.message.Message):
         description: builtins.str = ...,
         vector_index: builtins.bool = ...,
         vector_search_metric: builtins.str = ...,
+        vector_len: builtins.int = ...,
     ) -> None: ...
-    def ClearField(self, field_name: typing_extensions.Literal["description", b"description", "name", b"name", "tags", b"tags", "value_type", b"value_type", "vector_index", b"vector_index", "vector_search_metric", b"vector_search_metric"]) -> None: ...
+    def ClearField(self, field_name: typing_extensions.Literal["description", b"description", "name", b"name", "tags", b"tags", "value_type", b"value_type", "vector_index", b"vector_index", "vector_len", b"vector_len", "vector_search_metric", b"vector_search_metric"]) -> None: ...
 
 global___FeatureSpecV2 = FeatureSpecV2
@@ -68,7 +68,6 @@ def create_online_store(self) -> Dict[str, Any]:
             "password": "test!@#$%",
             "database": "test",
             "vector_enabled": True,
-            "vector_len": 2,
             "port": self.container.get_exposed_port(5432),
         }