wip: working multivector query

Author: justin-cechmanek

redisvl/query/aggregate.py

@@ -260,76 +260,139 @@ class MultiVectorQuery(AggregationQuery):
             results = index.query(query)
 
 
+
+        FT.AGGREGATE multi_vector_test 
+        "@user_embedding:[VECTOR_RANGE 2.0 $vector_0]=>{$YIELD_DISTANCE_AS: distance_0}
+        | @image_embedding:[VECTOR_RANGE 2.0 $vector_1]=>{$YIELD_DISTANCE_AS: distance_1}" 
+        PARAMS 4
+        vector_0 "\xcd\xcc\xcc=\xcd\xcc\xcc=\x00\x00\x00?" 
+        vector_1 "\x9a\x99\x99\x99\x99\x99\xb9?\x9a\x99\x99\x99\x99\x99\xb9?\x9a\x99\x99\x99\x99\x99\xb9?\x9a\x99\x99\x99\x99\x99\xb9?\x9a\x99\x99\x99\x99\x99\xb9?" 
+        APPLY "(2 - @distance_0)/2" AS score_0
+        APPLY "(2 - @distance_1)/2" AS score_1 
+        DIALECT 2
+        APPLY "(@score_0 + @score_1)" AS combined_score
+        SORTBY 2 @combined_score 
+        ASC 
+        MAX 10 
+        LOAD 2 score_0 score_1
+
+
+
+
+
     FT.AGGREGATE 'idx:characters'
-     "@embedding1:[VECTOR_RANGE .7 $vector1]=>{$YIELD_DISTANCE_AS: vector_distance1} | @embedding2:[VECTOR_RANGE 1.0 $vector2]=>{$YIELD_DISTANCE_AS: vector_distance2} | @embedding3:[VECTOR_RANGE 1.7 $vector3]=>{$YIELD_DISTANCE_AS: vector_distance3}  | @name:(James)"
-     ADDSCORES
-     SCORER BM25STD.NORM
-     LOAD 2 created_at @embedding
-     APPLY 'case(exists(@vector_distance1), @vector_distance1, 0.0)' as v1
-     APPLY 'case(exists(@vector_distance2), @vector_distance2, 0.0)' as v2
-     APPLY 'case(exists(@vector_distance3), @vector_distance3, 0.0)' as v3
+     "@embedding1:[VECTOR_RANGE .7 $vector1]=>{$YIELD_DISTANCE_AS: vector_distance1}
+     | @embedding2:[VECTOR_RANGE 1.0 $vector2]=>{$YIELD_DISTANCE_AS: vector_distance2}
+     | @embedding3:[VECTOR_RANGE 1.7 $vector3]=>{$YIELD_DISTANCE_AS: vector_distance3}
+     | @name:(James)"
+     ### ADDSCORES
+     ### SCORER BM25STD.NORM
+     ### LOAD 2 created_at @embedding
+     APPLY '(2 - @vector_distance1)/2' as v1
+     APPLY '(2 - @vector_distance2)/2' as v2
+     APPLY '(2 - @vector_distance3)/2' as v3
      APPLY '(@__score * 0.3 + (@v1 * 0.3) + (@v2 * 1.2) + (@v3 * 0.1))' AS final_score
      PARAMS 6 vector1 "\xe4\xd6..." vector2 "\x89\xa0..." vector3 "\x3c\x19..."
      SORTBY 2 @final_score DESC
      DIALECT 2
      LIMIT 0 100
 
-
     """
 
     DISTANCE_ID: str = "vector_distance"
-    VECTOR_PARAM: str = "vector"
 
     def __init__(
         self,
         vectors: Union[bytes, List[bytes], List[float], List[List[float]]],
         vector_field_names: Union[str, List[str]],
+        weights: List[float] = [1.0],
+        return_fields: Optional[List[str]] = None,
         filter_expression: Optional[Union[str, FilterExpression]] = None,
-        weights: Union[float, List[float]] = 1.0,
-        dtypes: Union[str, List[str]] = "float32",
+        dtypes: List[str] = ["float32"],
         num_results: int = 10,
-        return_fields: Optional[List[str]] = None,
+        return_score: bool = False,
         dialect: int = 2,
     ):
         """
         Instantiates a MultiVectorQuery object.
 
         Args:
             vectors (Union[bytes, List[bytes], List[float], List[List[float]]): The vectors to perform vector similarity search.
-            vector_field_names (str): The vector field names to search in.
-            filter_expression (Optional[FilterExpression], optional): The filter expression to use.
-                Defaults to None.
-            weights (Union[float, List[float]], optional): The weights of the vector similarity.
+            vector_field_names (Union[str, List[str]]): The vector field names to search in.
+            weights (List[float]): The weights of the vector similarity.
                 Documents will be scored as:
                 score = (w1) * score1 + (w2) * score2 + (w3) * score3 + ...
-                Defaults to 1.0, which corresponds to equal weighting
-            dtype (Union[str, List[str]] optional): The data types of the vectors. Defaults to "float32" for all vectors.
-            num_results (int, optional): The number of results to return. Defaults to 10.
+                Defaults to [1.0], which corresponds to equal weighting
             return_fields (Optional[List[str]], optional): The fields to return. Defaults to None.
+            filter_expression (Optional[Union[str, FilterExpression]]): The filter expression to use.
+                Defaults to None.
+            dtypes (List[str]): The data types of the vectors. Defaults to ["float32"] for all vectors.
+            num_results (int, optional): The number of results to return. Defaults to 10.
+            return_score (bool): Whether to return the combined vector similarity score.
+                Defaults to False.
             dialect (int, optional): The Redis dialect version. Defaults to 2.
 
         Raises:
             ValueError: The number of vectors, vector field names, and weights do not agree.
-            TypeError: If the stopwords are not a set, list, or tuple of strings.
         """
 
-        self._vectors = vectors
-        self._vector_fields = vector_field_names
         self._filter_expression = filter_expression
-        self._weights = weights
         self._dtypes = dtypes
         self._num_results = num_results
 
+        if len(vectors) == 0 or len(vector_field_names) == 0 or len(weights) == 0:
+            raise ValueError(
+                f"""The number of vectors and vector field names must be equal.
+                             If weights are specified their number must match the number of vectors and vector field names also.
+                            Length of vectors list: {len(vectors) = }
+                            Length of vector_field_names list: {len(vector_field_names) = }
+                            Length of weights list: {len(weights) = }
+                            """
+            )
+
+        if isinstance(vectors, bytes) or isinstance(vectors[0], float):
+            self._vectors = [vectors]
+        else:
+            self._vectors = vectors
+        if isinstance(vector_field_names, str):
+            self._vector_field_names = [vector_field_names]
+        else:
+            self._vector_field_names = vector_field_names
+        if len(weights) == 1:
+            self._weights = weights * len(vectors)
+        else:
+            self._weights = weights
+        if len(dtypes) == 1:
+            self._dtypes = dtypes * len(vectors)
+        else:
+            self._dtypes = dtypes
+
+        if (len(self._vectors) != len(self._vector_field_names)) or (
+            len(self._vectors) != len(self._weights)
+        ):
+            raise ValueError(
+                f"""The number of vectors and vector field names must be equal.
+                             If weights are specified their number must match the number of vectors and vector field names also.
+                            Length of vectors list: {len(self._vectors) = }
+                            Length of vector_field_names list: {len(self._vector_field_names) = }
+                            Length of weights list: {len(self._weights) = }
+                            """
+            )
+
         query_string = self._build_query_string()
         super().__init__(query_string)
 
-        self.scorer(text_scorer)
-        self.add_scores()
-        self.apply(
-            vector_similarity=f"(2 - @{self.DISTANCE_ID})/2", text_score="@__score"
-        )
-        self.apply(hybrid_score=f"{1-alpha}*@text_score + {alpha}*@vector_similarity")
-        self.sort_by(Desc("@hybrid_score"), max=num_results)  # type: ignore
+        # construct the scoring string based on the vector similarity scores and weights
+        combined_scores = []
+        for i, w in enumerate(self._weights):
+            combined_scores.append(f"@score_{i} * {w}")
+        combined_score_string = " + ".join(combined_scores)
+        combined_score_string = f"'({combined_score_string})'"
+
+        self.apply(combined_score=combined_score_string)
+
+        # self.add_scores()
+        self.sort_by(Desc("@combined_score"), max=num_results)  # type: ignore
         self.dialect(dialect)
         if return_fields:
             self.load(*return_fields)  # type: ignore[arg-type]
@@ -341,49 +404,45 @@ def params(self) -> Dict[str, Any]:
         Returns:
             Dict[str, Any]: The parameters for the aggregation.
         """
-        if isinstance(self._vector, list):
-            vector = array_to_buffer(self._vector, dtype=self._dtype)
-        else:
-            vector = self._vector
-
-        params = {self.VECTOR_PARAM: vector}
-
+        params = {}
+        for i, (vector, vector_field, dtype) in enumerate(zip(
+            self._vectors, self._vector_field_names, self._dtypes
+        )):
+            if isinstance(vector, list):
+                vector = array_to_buffer(vector, dtype=dtype)
+            params[f"vector_{i}"] = vector
         return params
 
-    def _tokenize_and_escape_query(self, user_query: str) -> str:
-        """Convert a raw user query to a redis full text query joined by ORs
-        Args:
-            user_query (str): The user query to tokenize and escape.
-
-        Returns:
-            str: The tokenized and escaped query string.
-        Raises:
-            ValueError: If the text string becomes empty after stopwords are removed.
-        """
-        escaper = TokenEscaper()
-
-        tokens = [
-            escaper.escape(
-                token.strip().strip(",").replace("“", "").replace("”", "").lower()
-            )
-            for token in user_query.split()
-        ]
-        tokenized = " | ".join(
-            [token for token in tokens if token and token not in self._stopwords]
-        )
-
-        if not tokenized:
-            raise ValueError("text string cannot be empty after removing stopwords")
-        return tokenized
-
     def _build_query_string(self) -> str:
         """Build the full query string for text search with optional filtering."""
+
+        filter_expression = self._filter_expression
         if isinstance(self._filter_expression, FilterExpression):
             filter_expression = str(self._filter_expression)
-        else:
-            filter_expression = ""
 
         # base KNN query
-        knn_query = f"KNN {self._num_results} @{self._vector_field} ${self.VECTOR_PARAM} AS {self.DISTANCE_ID}"
+        knn_queries = []
+        range_queries = []
+        for i, (vector, field) in enumerate(zip(self._vectors, self._vector_field_names)):
+            knn_queries.append(f"[KNN {self._num_results} @{field} $vector_{i} AS distance_{i}]")
+            range_queries.append(f"@{field}:[VECTOR_RANGE 2.0 $vector_{i}]=>{{$YIELD_DISTANCE_AS: distance_{i}}}")
+
+        knn_query = " | ".join(knn_queries) ## knn_queries format doesn't work
+        knn_query = " | ".join(range_queries)
+
+        # calculate the respective vector similarities
+        apply_string = ""
+        for i, (vector, field_name, weight) in enumerate(
+            zip(self._vectors, self._vector_field_names, self._weights)
+        ):
+            apply_string += f'APPLY "(2 - @distance_{i})/2" AS score_{i} '
 
-        return f"{filter_expression})=>[{knn_query}]"
+        return (
+            f"{knn_query} {filter_expression} {apply_string}"
+            if filter_expression
+            else f"{knn_query} {apply_string}"
+        )
+
+    def __str__(self) -> str:
+        """Return the string representation of the query."""
+        return " ".join([str(x) for x in self.build_args()])