qdrant-haystack 4.0.0__tar.gz → 4.1.1__tar.gz

{qdrant_haystack-4.0.0 → qdrant_haystack-4.1.1}/CHANGELOG.md

@@ -1,5 +1,24 @@
 # Changelog
 
+## [integrations/qdrant-v4.1.0] - 2024-07-03
+
+### 🚀 Features
+
+- Add `score_threshold` to Qdrant Retrievers (#860)
+- Qdrant - add support for BM42 (#864)
+
+## [integrations/qdrant-v4.0.0] - 2024-07-02
+
+### 🚜 Refactor
+
+- [**breaking**] Qdrant - remove unused init parameters: `content_field`, `name_field`, `embedding_field`, and `duplicate_documents` (#861)
+- [**breaking**] Qdrant - set `scale_score` default value to `False` (#862)
+
+### ⚙️ Miscellaneous Tasks
+
+- Retry tests to reduce flakyness (#836)
+- Update ruff invocation to include check parameter (#853)
+
 ## [integrations/qdrant-v3.8.1] - 2024-06-20
 
 ### 📚 Documentation

{qdrant_haystack-4.0.0 → qdrant_haystack-4.1.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: qdrant-haystack
-Version: 4.0.0
+Version: 4.1.1
 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
@@ -18,8 +18,8 @@ Classifier: Programming Language :: Python :: 3.11
 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: haystack-ai
+Requires-Dist: qdrant-client>=1.10.0
 Description-Content-Type: text/markdown
 
 # qdrant-haystack

{qdrant_haystack-4.0.0 → qdrant_haystack-4.1.1}/pyproject.toml

@@ -25,7 +25,7 @@ classifiers = [
   "Programming Language :: Python :: Implementation :: CPython",
   "Programming Language :: Python :: Implementation :: PyPy",
 ]
-dependencies = ["haystack-ai>=2.0.1", "qdrant-client"]
+dependencies = ["haystack-ai", "qdrant-client>=1.10.0"]
 
 [project.urls]
 Source = "https://github.com/deepset-ai/haystack-core-integrations"

{qdrant_haystack-4.0.0 → qdrant_haystack-4.1.1}/src/haystack_integrations/components/retrievers/qdrant/retriever.py

@@ -2,6 +2,8 @@ from typing import Any, Dict, List, Optional, Union
 
 from haystack import Document, component, default_from_dict, default_to_dict
 from haystack.dataclasses.sparse_embedding import SparseEmbedding
+from haystack.document_stores.types import FilterPolicy
+from haystack.document_stores.types.filter_policy import apply_filter_policy
 from haystack_integrations.document_stores.qdrant import QdrantDocumentStore
 from qdrant_client.http import models
 
@@ -39,6 +41,8 @@ class QdrantEmbeddingRetriever:
         top_k: int = 10,
         scale_score: bool = False,
         return_embedding: bool = False,
+        filter_policy: Union[str, FilterPolicy] = FilterPolicy.REPLACE,
+        score_threshold: Optional[float] = None,
     ):
         """
         Create a QdrantEmbeddingRetriever component.
@@ -48,6 +52,11 @@ 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 filter_policy: Policy to determine how filters are applied.
+        :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 +70,10 @@ class QdrantEmbeddingRetriever:
         self._top_k = top_k
         self._scale_score = scale_score
         self._return_embedding = return_embedding
+        self._filter_policy = (
+            filter_policy if isinstance(filter_policy, FilterPolicy) else FilterPolicy.from_str(filter_policy)
+        )
+        self._score_threshold = score_threshold
 
     def to_dict(self) -> Dict[str, Any]:
         """
@@ -74,8 +87,10 @@ class QdrantEmbeddingRetriever:
             document_store=self._document_store,
             filters=self._filters,
             top_k=self._top_k,
+            filter_policy=self._filter_policy.value,
             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()
 
@@ -93,6 +108,7 @@ class QdrantEmbeddingRetriever:
         """
         document_store = QdrantDocumentStore.from_dict(data["init_parameters"]["document_store"])
         data["init_parameters"]["document_store"] = document_store
+        data["init_parameters"]["filter_policy"] = FilterPolicy.from_str(data["init_parameters"]["filter_policy"])
         return default_from_dict(cls, data)
 
     @component.output_types(documents=List[Document])
@@ -103,6 +119,7 @@ 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.
@@ -112,16 +129,20 @@ 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.
 
         """
+        filters = apply_filter_policy(self._filter_policy, self._filters, filters)
+
         docs = self._document_store._query_by_embedding(
             query_embedding=query_embedding,
-            filters=filters or self._filters,
+            filters=filters,
             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 +182,8 @@ class QdrantSparseEmbeddingRetriever:
         top_k: int = 10,
         scale_score: bool = False,
         return_embedding: bool = False,
+        filter_policy: Union[str, FilterPolicy] = FilterPolicy.REPLACE,
+        score_threshold: Optional[float] = None,
     ):
         """
         Create a QdrantSparseEmbeddingRetriever component.
@@ -170,6 +193,11 @@ 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 filter_policy: Policy to determine how filters are applied. Defaults to "replace".
+        :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 +211,10 @@ class QdrantSparseEmbeddingRetriever:
         self._top_k = top_k
         self._scale_score = scale_score
         self._return_embedding = return_embedding
+        self._filter_policy = (
+            filter_policy if isinstance(filter_policy, FilterPolicy) else FilterPolicy.from_str(filter_policy)
+        )
+        self._score_threshold = score_threshold
 
     def to_dict(self) -> Dict[str, Any]:
         """
@@ -197,7 +229,9 @@ class QdrantSparseEmbeddingRetriever:
             filters=self._filters,
             top_k=self._top_k,
             scale_score=self._scale_score,
+            filter_policy=self._filter_policy.value,
             return_embedding=self._return_embedding,
+            score_threshold=self._score_threshold,
         )
         d["init_parameters"]["document_store"] = self._document_store.to_dict()
 
@@ -215,6 +249,7 @@ class QdrantSparseEmbeddingRetriever:
         """
         document_store = QdrantDocumentStore.from_dict(data["init_parameters"]["document_store"])
         data["init_parameters"]["document_store"] = document_store
+        data["init_parameters"]["filter_policy"] = FilterPolicy.from_str(data["init_parameters"]["filter_policy"])
         return default_from_dict(cls, data)
 
     @component.output_types(documents=List[Document])
@@ -225,25 +260,35 @@ 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.
 
         :param query_sparse_embedding: Sparse Embedding of the query.
-        :param filters: A dictionary with filters to narrow down the search space.
+        :param filters: Filters applied to the retrieved Documents. The way runtime filters are applied depends on
+                        the `filter_policy` chosen at retriever initialization. See init method docstring for more
+                        details.
         :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.
 
         """
+        filters = apply_filter_policy(self._filter_policy, self._filters, filters)
+
         docs = self._document_store._query_by_sparse(
             query_sparse_embedding=query_sparse_embedding,
-            filters=filters or self._filters,
+            filters=filters,
             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 +333,8 @@ class QdrantHybridRetriever:
         filters: Optional[Union[Dict[str, Any], models.Filter]] = None,
         top_k: int = 10,
         return_embedding: bool = False,
+        filter_policy: Union[str, FilterPolicy] = FilterPolicy.REPLACE,
+        score_threshold: Optional[float] = None,
     ):
         """
         Create a QdrantHybridRetriever component.
@@ -296,6 +343,11 @@ 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 filter_policy: Policy to determine how filters are applied.
+        :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 +360,10 @@ class QdrantHybridRetriever:
         self._filters = filters
         self._top_k = top_k
         self._return_embedding = return_embedding
+        self._filter_policy = (
+            filter_policy if isinstance(filter_policy, FilterPolicy) else FilterPolicy.from_str(filter_policy)
+        )
+        self._score_threshold = score_threshold
 
     def to_dict(self) -> Dict[str, Any]:
         """
@@ -321,7 +377,9 @@ class QdrantHybridRetriever:
             document_store=self._document_store.to_dict(),
             filters=self._filters,
             top_k=self._top_k,
+            filter_policy=self._filter_policy.value,
             return_embedding=self._return_embedding,
+            score_threshold=self._score_threshold,
         )
 
     @classmethod
@@ -336,6 +394,7 @@ class QdrantHybridRetriever:
         """
         document_store = QdrantDocumentStore.from_dict(data["init_parameters"]["document_store"])
         data["init_parameters"]["document_store"] = document_store
+        data["init_parameters"]["filter_policy"] = FilterPolicy.from_str(data["init_parameters"]["filter_policy"])
         return default_from_dict(cls, data)
 
     @component.output_types(documents=List[Document])
@@ -346,25 +405,35 @@ 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.
 
         :param query_embedding: Dense embedding of the query.
         :param query_sparse_embedding: Sparse embedding of the query.
-        :param filters: A dictionary with filters to narrow down the search space.
+        :param filters: Filters applied to the retrieved Documents. The way runtime filters are applied depends on
+                        the `filter_policy` chosen at retriever initialization. See init method docstring for more
+                        details.
         :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.
 
         """
+        filters = apply_filter_policy(self._filter_policy, self._filters, filters)
+
         docs = self._document_store._query_hybrid(
             query_embedding=query_embedding,
             query_sparse_embedding=query_sparse_embedding,
-            filters=filters or self._filters,
+            filters=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}

{qdrant_haystack-4.0.0 → qdrant_haystack-4.1.1}/src/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,
                 ),
             }