Support dense vectors based on numpy arrays

Author: miguelgrinberg

elasticsearch/dsl/field.py

@@ -1616,11 +1616,11 @@ def __init__(
             kwargs["multi"] = True
         super().__init__(*args, **kwargs)
 
-    def _deserialize(self, data: Any) -> Any:
-        if self._element_type == "float":
-            return float(data)
-        elif self._element_type == "byte":
-            return int(data)
+    def clean(self, data: Any) -> Any:
+        if data is not None:
+            data = self.deserialize(data)
+        if (data is None or len(data) == 0) and self._required:
+            raise ValidationException("Value required for this field.")
         return data
 
 

elasticsearch/dsl/utils.py

@@ -612,7 +612,9 @@ def to_dict(self, skip_empty: bool = True) -> Dict[str, Any]:
             if skip_empty:
                 # don't serialize empty values
                 # careful not to include numeric zeros
-                if v in ([], {}, None):
+                # the "is" operator is used below because it is the only comparison
+                # that works for numpy arrays
+                if v is [] or v is {} or v is None:
                     continue
 
             out[k] = v

examples/quotes/backend/quotes.py

@@ -135,7 +135,7 @@ async def search_quotes(req: SearchRequest) -> SearchResponse:
 def embed_quotes(quotes):
     embeddings = model.encode([q.quote for q in quotes])
     for q, e in zip(quotes, embeddings):
-        q.embedding = e.tolist()
+        q.embedding = e
 
 
 async def ingest_quotes():

utils/templates/field.py.tpl

@@ -418,11 +418,11 @@ class {{ k.name }}({{ k.parent }}):
             kwargs["multi"] = True
         super().__init__(*args, **kwargs)
 
-    def _deserialize(self, data: Any) -> Any:
-        if self._element_type == "float":
-            return float(data)
-        elif self._element_type == "byte":
-            return int(data)
+    def clean(self, data: Any) -> Any:
+        if data is not None:
+            data = self.deserialize(data)
+        if (data is None or len(data) == 0) and self._required:
+            raise ValidationException("Value required for this field.")
         return data
         {% elif k.field == "scaled_float" %}
         if 'scaling_factor' not in kwargs: