qdrant-haystack 4.0.0__py3-none-any.whl → 4.1.0__py3-none-any.whl

haystack_integrations/components/retrievers/qdrant/retriever.py

@@ -39,6 +39,7 @@ class QdrantEmbeddingRetriever:
         top_k: int = 10,
         scale_score: bool = False,
         return_embedding: bool = False,
+        score_threshold: Optional[float] = None,
     ):
         """
         Create a QdrantEmbeddingRetriever component.
@@ -48,6 +49,10 @@ class QdrantEmbeddingRetriever:
         :param top_k: The maximum number of documents to retrieve.
         :param scale_score: Whether to scale the scores of the retrieved documents or not.
         :param return_embedding: Whether to return the embedding of the retrieved Documents.
+        :param score_threshold: A minimal score threshold for the result.
+            Score of the returned result might be higher or smaller than the threshold
+             depending on the `similarity` function specified in the Document Store.
+            E.g. for cosine similarity only higher scores will be returned.
 
         :raises ValueError: If `document_store` is not an instance of `QdrantDocumentStore`.
         """
@@ -61,6 +66,7 @@ class QdrantEmbeddingRetriever:
         self._top_k = top_k
         self._scale_score = scale_score
         self._return_embedding = return_embedding
+        self._score_threshold = score_threshold
 
     def to_dict(self) -> Dict[str, Any]:
         """
@@ -76,6 +82,7 @@ class QdrantEmbeddingRetriever:
             top_k=self._top_k,
             scale_score=self._scale_score,
             return_embedding=self._return_embedding,
+            score_threshold=self._score_threshold,
         )
         d["init_parameters"]["document_store"] = self._document_store.to_dict()
 
@@ -103,6 +110,7 @@ class QdrantEmbeddingRetriever:
         top_k: Optional[int] = None,
         scale_score: Optional[bool] = None,
         return_embedding: Optional[bool] = None,
+        score_threshold: Optional[float] = None,
     ):
         """
         Run the Embedding Retriever on the given input data.
@@ -112,6 +120,7 @@ class QdrantEmbeddingRetriever:
         :param top_k: The maximum number of documents to return.
         :param scale_score: Whether to scale the scores of the retrieved documents or not.
         :param return_embedding: Whether to return the embedding of the retrieved Documents.
+        :param score_threshold: A minimal score threshold for the result.
         :returns:
             The retrieved documents.
 
@@ -122,6 +131,7 @@ class QdrantEmbeddingRetriever:
             top_k=top_k or self._top_k,
             scale_score=scale_score or self._scale_score,
             return_embedding=return_embedding or self._return_embedding,
+            score_threshold=score_threshold or self._score_threshold,
         )
 
         return {"documents": docs}
@@ -161,6 +171,7 @@ class QdrantSparseEmbeddingRetriever:
         top_k: int = 10,
         scale_score: bool = False,
         return_embedding: bool = False,
+        score_threshold: Optional[float] = None,
     ):
         """
         Create a QdrantSparseEmbeddingRetriever component.
@@ -170,6 +181,10 @@ class QdrantSparseEmbeddingRetriever:
         :param top_k: The maximum number of documents to retrieve.
         :param scale_score: Whether to scale the scores of the retrieved documents or not.
         :param return_embedding: Whether to return the sparse embedding of the retrieved Documents.
+        :param score_threshold: A minimal score threshold for the result.
+            Score of the returned result might be higher or smaller than the threshold
+             depending on the Distance function used.
+            E.g. for cosine similarity only higher scores will be returned.
 
         :raises ValueError: If `document_store` is not an instance of `QdrantDocumentStore`.
         """
@@ -183,6 +198,7 @@ class QdrantSparseEmbeddingRetriever:
         self._top_k = top_k
         self._scale_score = scale_score
         self._return_embedding = return_embedding
+        self._score_threshold = score_threshold
 
     def to_dict(self) -> Dict[str, Any]:
         """
@@ -198,6 +214,7 @@ class QdrantSparseEmbeddingRetriever:
             top_k=self._top_k,
             scale_score=self._scale_score,
             return_embedding=self._return_embedding,
+            score_threshold=self._score_threshold,
         )
         d["init_parameters"]["document_store"] = self._document_store.to_dict()
 
@@ -225,6 +242,7 @@ class QdrantSparseEmbeddingRetriever:
         top_k: Optional[int] = None,
         scale_score: Optional[bool] = None,
         return_embedding: Optional[bool] = None,
+        score_threshold: Optional[float] = None,
     ):
         """
         Run the Sparse Embedding Retriever on the given input data.
@@ -234,6 +252,10 @@ class QdrantSparseEmbeddingRetriever:
         :param top_k: The maximum number of documents to return.
         :param scale_score: Whether to scale the scores of the retrieved documents or not.
         :param return_embedding: Whether to return the embedding of the retrieved Documents.
+        :param score_threshold: A minimal score threshold for the result.
+            Score of the returned result might be higher or smaller than the threshold
+             depending on the Distance function used.
+            E.g. for cosine similarity only higher scores will be returned.
         :returns:
             The retrieved documents.
 
@@ -244,6 +266,7 @@ class QdrantSparseEmbeddingRetriever:
             top_k=top_k or self._top_k,
             scale_score=scale_score or self._scale_score,
             return_embedding=return_embedding or self._return_embedding,
+            score_threshold=score_threshold or self._score_threshold,
         )
 
         return {"documents": docs}
@@ -288,6 +311,7 @@ class QdrantHybridRetriever:
         filters: Optional[Union[Dict[str, Any], models.Filter]] = None,
         top_k: int = 10,
         return_embedding: bool = False,
+        score_threshold: Optional[float] = None,
     ):
         """
         Create a QdrantHybridRetriever component.
@@ -296,6 +320,10 @@ class QdrantHybridRetriever:
         :param filters: A dictionary with filters to narrow down the search space.
         :param top_k: The maximum number of documents to retrieve.
         :param return_embedding: Whether to return the embeddings of the retrieved Documents.
+        :param score_threshold: A minimal score threshold for the result.
+            Score of the returned result might be higher or smaller than the threshold
+             depending on the Distance function used.
+            E.g. for cosine similarity only higher scores will be returned.
 
         :raises ValueError: If 'document_store' is not an instance of QdrantDocumentStore.
         """
@@ -308,6 +336,7 @@ class QdrantHybridRetriever:
         self._filters = filters
         self._top_k = top_k
         self._return_embedding = return_embedding
+        self._score_threshold = score_threshold
 
     def to_dict(self) -> Dict[str, Any]:
         """
@@ -322,6 +351,7 @@ class QdrantHybridRetriever:
             filters=self._filters,
             top_k=self._top_k,
             return_embedding=self._return_embedding,
+            score_threshold=self._score_threshold,
         )
 
     @classmethod
@@ -346,6 +376,7 @@ class QdrantHybridRetriever:
         filters: Optional[Union[Dict[str, Any], models.Filter]] = None,
         top_k: Optional[int] = None,
         return_embedding: Optional[bool] = None,
+        score_threshold: Optional[float] = None,
     ):
         """
         Run the Sparse Embedding Retriever on the given input data.
@@ -355,6 +386,10 @@ class QdrantHybridRetriever:
         :param filters: A dictionary with filters to narrow down the search space.
         :param top_k: The maximum number of documents to return.
         :param return_embedding: Whether to return the embedding of the retrieved Documents.
+        :param score_threshold: A minimal score threshold for the result.
+            Score of the returned result might be higher or smaller than the threshold
+             depending on the Distance function used.
+            E.g. for cosine similarity only higher scores will be returned.
         :returns:
             The retrieved documents.
 
@@ -365,6 +400,7 @@ class QdrantHybridRetriever:
             filters=filters or self._filters,
             top_k=top_k or self._top_k,
             return_embedding=return_embedding or self._return_embedding,
+            score_threshold=score_threshold or self._score_threshold,
         )
 
         return {"documents": docs}

haystack_integrations/document_stores/qdrant/document_store.py

@@ -111,6 +111,7 @@ class QdrantDocumentStore:
         embedding_dim: int = 768,
         on_disk: bool = False,
         use_sparse_embeddings: bool = False,
+        sparse_idf: bool = False,
         similarity: str = "cosine",
         return_embedding: bool = False,
         progress_bar: bool = True,
@@ -168,6 +169,9 @@ class QdrantDocumentStore:
             Whether to store the collection on disk.
         :param use_sparse_embedding:
             If set to `True`, enables support for sparse embeddings.
+        :param sparse_idf:
+            If set to `True`, computes the Inverse Document Frequency (IDF) when using sparse embeddings.
+            It is required to use techniques like BM42. It is ignored if `use_sparse_embeddings` is `False`.
         :param similarity:
             The similarity metric to use.
         :param return_embedding:
@@ -246,6 +250,7 @@ class QdrantDocumentStore:
         self.recreate_index = recreate_index
         self.payload_fields_to_index = payload_fields_to_index
         self.use_sparse_embeddings = use_sparse_embeddings
+        self.sparse_idf = use_sparse_embeddings and sparse_idf
         self.embedding_dim = embedding_dim
         self.on_disk = on_disk
         self.similarity = similarity
@@ -280,6 +285,7 @@ class QdrantDocumentStore:
                 self.recreate_index,
                 self.similarity,
                 self.use_sparse_embeddings,
+                self.sparse_idf,
                 self.on_disk,
                 self.payload_fields_to_index,
             )
@@ -347,7 +353,9 @@ class QdrantDocumentStore:
             if not isinstance(doc, Document):
                 msg = f"DocumentStore.write_documents() expects a list of Documents but got an element of {type(doc)}."
                 raise ValueError(msg)
-        self._set_up_collection(self.index, self.embedding_dim, False, self.similarity, self.use_sparse_embeddings)
+        self._set_up_collection(
+            self.index, self.embedding_dim, False, self.similarity, self.use_sparse_embeddings, self.sparse_idf
+        )
 
         if len(documents) == 0:
             logger.warning("Calling QdrantDocumentStore.write_documents() with empty list")
@@ -498,6 +506,7 @@ class QdrantDocumentStore:
         top_k: int = 10,
         scale_score: bool = False,
         return_embedding: bool = False,
+        score_threshold: Optional[float] = None,
     ) -> List[Document]:
         """
         Queries Qdrant using a sparse embedding and returns the most relevant documents.
@@ -507,6 +516,10 @@ class QdrantDocumentStore:
         :param top_k: Maximum number of documents to return.
         :param scale_score: Whether to scale the scores of the retrieved documents.
         :param return_embedding: Whether to return the embeddings of the retrieved documents.
+        :param score_threshold: A minimal score threshold for the result.
+            Score of the returned result might be higher or smaller than the threshold
+             depending on the Distance function used.
+            E.g. for cosine similarity only higher scores will be returned.
 
         :returns: List of documents that are most similar to `query_sparse_embedding`.
 
@@ -536,6 +549,7 @@ class QdrantDocumentStore:
             query_filter=qdrant_filters,
             limit=top_k,
             with_vectors=return_embedding,
+            score_threshold=score_threshold,
         )
         results = [
             convert_qdrant_point_to_haystack_document(point, use_sparse_embeddings=self.use_sparse_embeddings)
@@ -555,6 +569,7 @@ class QdrantDocumentStore:
         top_k: int = 10,
         scale_score: bool = False,
         return_embedding: bool = False,
+        score_threshold: Optional[float] = None,
     ) -> List[Document]:
         """
         Queries Qdrant using a dense embedding and returns the most relevant documents.
@@ -564,6 +579,10 @@ class QdrantDocumentStore:
         :param top_k: Maximum number of documents to return.
         :param scale_score: Whether to scale the scores of the retrieved documents.
         :param return_embedding: Whether to return the embeddings of the retrieved documents.
+        :param score_threshold: A minimal score threshold for the result.
+            Score of the returned result might be higher or smaller than the threshold
+             depending on the Distance function used.
+            E.g. for cosine similarity only higher scores will be returned.
 
         :returns: List of documents that are most similar to `query_embedding`.
         """
@@ -578,6 +597,7 @@ class QdrantDocumentStore:
             query_filter=qdrant_filters,
             limit=top_k,
             with_vectors=return_embedding,
+            score_threshold=score_threshold,
         )
         results = [
             convert_qdrant_point_to_haystack_document(point, use_sparse_embeddings=self.use_sparse_embeddings)
@@ -600,6 +620,7 @@ class QdrantDocumentStore:
         filters: Optional[Union[Dict[str, Any], rest.Filter]] = None,
         top_k: int = 10,
         return_embedding: bool = False,
+        score_threshold: Optional[float] = None,
     ) -> List[Document]:
         """
         Retrieves documents based on dense and sparse embeddings and fuses the results using Reciprocal Rank Fusion.
@@ -612,6 +633,10 @@ class QdrantDocumentStore:
         :param filters: Filters applied to the retrieved documents.
         :param top_k: Maximum number of documents to return.
         :param return_embedding: Whether to return the embeddings of the retrieved documents.
+        :param score_threshold: A minimal score threshold for the result.
+            Score of the returned result might be higher or smaller than the threshold
+             depending on the Distance function used.
+            E.g. for cosine similarity only higher scores will be returned.
 
         :returns: List of Document that are most similar to `query_embedding` and `query_sparse_embedding`.
 
@@ -642,6 +667,7 @@ class QdrantDocumentStore:
             limit=top_k,
             with_payload=True,
             with_vector=return_embedding,
+            score_threshold=score_threshold,
         )
 
         dense_request = rest.SearchRequest(
@@ -714,6 +740,7 @@ class QdrantDocumentStore:
         recreate_collection: bool,
         similarity: str,
         use_sparse_embeddings: bool,
+        sparse_idf: bool,
         on_disk: bool = False,
         payload_fields_to_index: Optional[List[dict]] = None,
     ):
@@ -729,6 +756,8 @@ class QdrantDocumentStore:
             The similarity measure to use.
         :param use_sparse_embeddings:
             Whether to use sparse embeddings.
+        :param sparse_idf:
+            Whether to compute the Inverse Document Frequency (IDF) when using sparse embeddings. Required for BM42.
         :param on_disk:
             Whether to store the collection on disk.
         :param payload_fields_to_index:
@@ -745,7 +774,9 @@ class QdrantDocumentStore:
         if recreate_collection or not self.client.collection_exists(collection_name):
             # There is no need to verify the current configuration of that
             # collection. It might be just recreated again or does not exist yet.
-            self.recreate_collection(collection_name, distance, embedding_dim, on_disk, use_sparse_embeddings)
+            self.recreate_collection(
+                collection_name, distance, embedding_dim, on_disk, use_sparse_embeddings, sparse_idf
+            )
             # Create Payload index if payload_fields_to_index is provided
             self._create_payload_index(collection_name, payload_fields_to_index)
             return
@@ -808,6 +839,7 @@ class QdrantDocumentStore:
         embedding_dim: int,
         on_disk: Optional[bool] = None,
         use_sparse_embeddings: Optional[bool] = None,
+        sparse_idf: bool = False,
     ):
         """
         Recreates the Qdrant collection with the specified parameters.
@@ -822,6 +854,8 @@ class QdrantDocumentStore:
             Whether to store the collection on disk.
         :param use_sparse_embeddings:
             Whether to use sparse embeddings.
+        :param sparse_idf:
+            Whether to compute the Inverse Document Frequency (IDF) when using sparse embeddings. Required for BM42.
         """
         if on_disk is None:
             on_disk = self.on_disk
@@ -840,7 +874,8 @@ class QdrantDocumentStore:
                 SPARSE_VECTORS_NAME: rest.SparseVectorParams(
                     index=rest.SparseIndexParams(
                         on_disk=on_disk,
-                    )
+                    ),
+                    modifier=rest.Modifier.IDF if sparse_idf else None,
                 ),
             }
 

{qdrant_haystack-4.0.0.dist-info → qdrant_haystack-4.1.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: qdrant-haystack
-Version: 4.0.0
+Version: 4.1.0
 Summary: An integration of Qdrant ANN vector database backend with Haystack
 Project-URL: Source, https://github.com/deepset-ai/haystack-core-integrations
 Project-URL: Documentation, https://github.com/deepset-ai/haystack-core-integrations/blob/main/integrations/qdrant/README.md
@@ -19,7 +19,7 @@ Classifier: Programming Language :: Python :: Implementation :: CPython
 Classifier: Programming Language :: Python :: Implementation :: PyPy
 Requires-Python: >=3.8
 Requires-Dist: haystack-ai>=2.0.1
-Requires-Dist: qdrant-client
+Requires-Dist: qdrant-client>=1.10.0
 Description-Content-Type: text/markdown
 
 # qdrant-haystack

{qdrant_haystack-4.0.0.dist-info → qdrant_haystack-4.1.0.dist-info}/RECORD

@@ -1,11 +1,11 @@
 haystack_integrations/components/retrievers/qdrant/__init__.py,sha256=IRjcM4f8b5eKFEMn8tn6h6RrfslEGP3WafU7mrzNzQM,313
-haystack_integrations/components/retrievers/qdrant/retriever.py,sha256=_6noYJ0M71shgoTOywIgSuGQtB-CBhwRW_zUFiYIOTw,13465
+haystack_integrations/components/retrievers/qdrant/retriever.py,sha256=acSMsrtVbkU1xpihH6oeIaKy6SGnQvzERZ3TWJvUViY,15826
 haystack_integrations/document_stores/qdrant/__init__.py,sha256=kUGc5uewqArhmVR-JqB_NmJ4kNkTIQIvYDNSoO2ELn0,302
 haystack_integrations/document_stores/qdrant/converters.py,sha256=2hcuI3kty1dVHzX1WGXxEtlrnZ9E8TAG56XATCFa6Pw,2491
-haystack_integrations/document_stores/qdrant/document_store.py,sha256=mjzv6Z3iE9oFRil_PVLjmEq-vX7a7ULpT5afGsU7iSU,36088
+haystack_integrations/document_stores/qdrant/document_store.py,sha256=eLw4P1h8GCj40R-BIlQOvJG9MpDzvtmQ7Hpb3AZhMSo,38117
 haystack_integrations/document_stores/qdrant/filters.py,sha256=0w70Wa3Za1fNdbJ5O95sZDIpXfblJG_sBBUv0JTQ0-o,8337
 haystack_integrations/document_stores/qdrant/migrate_to_sparse.py,sha256=i6wBC_9_JVzYZtqKm3dhHKTxhwNdcAdpgki8GABDp1c,4909
-qdrant_haystack-4.0.0.dist-info/METADATA,sha256=wHvVJIDCQDPFLX8fL_d11zNMZul4U6r02bVhhCdmitk,1862
-qdrant_haystack-4.0.0.dist-info/WHEEL,sha256=1yFddiXMmvYK7QYTqtRNtX66WJ0Mz8PYEiEUoOUUxRY,87
-qdrant_haystack-4.0.0.dist-info/licenses/LICENSE.txt,sha256=B05uMshqTA74s-0ltyHKI6yoPfJ3zYgQbvcXfDVGFf8,10280
-qdrant_haystack-4.0.0.dist-info/RECORD,,
+qdrant_haystack-4.1.0.dist-info/METADATA,sha256=wGNRbZYtVJ3BbHH22ZsoVhGGtyCR3cNeozFKHiAsYjI,1870
+qdrant_haystack-4.1.0.dist-info/WHEEL,sha256=1yFddiXMmvYK7QYTqtRNtX66WJ0Mz8PYEiEUoOUUxRY,87
+qdrant_haystack-4.1.0.dist-info/licenses/LICENSE.txt,sha256=B05uMshqTA74s-0ltyHKI6yoPfJ3zYgQbvcXfDVGFf8,10280
+qdrant_haystack-4.1.0.dist-info/RECORD,,