feat: introduce SparseEmbedding (#7382)

* introduce SparseEmbedding

* reno

* add to pydoc config

docs/pydoc/config/data_classess_api.yml

@@ -2,7 +2,7 @@ loaders:
   - type: haystack_pydoc_tools.loaders.CustomPythonLoader
     search_path: [../../../haystack/dataclasses]
     modules:
-      ["answer", "byte_stream", "chat_message", "document", "streaming_chunk"]
+      ["answer", "byte_stream", "chat_message", "document", "streaming_chunk", "sparse_embedding"]
     ignore_when_discovered: ["__init__"]
 processors:
   - type: filter

haystack/dataclasses/__init__.py

@@ -2,6 +2,7 @@
 from haystack.dataclasses.byte_stream import ByteStream
 from haystack.dataclasses.chat_message import ChatMessage, ChatRole
 from haystack.dataclasses.document import Document
+from haystack.dataclasses.sparse_embedding import SparseEmbedding
 from haystack.dataclasses.streaming_chunk import StreamingChunk
 
 __all__ = [
@@ -13,4 +14,5 @@
     "ChatMessage",
     "ChatRole",
     "StreamingChunk",
+    "SparseEmbedding",
 ]

haystack/dataclasses/document.py

@@ -8,6 +8,7 @@
 
 from haystack import logging
 from haystack.dataclasses.byte_stream import ByteStream
+from haystack.dataclasses.sparse_embedding import SparseEmbedding
 
 logger = logging.getLogger(__name__)
 
@@ -57,7 +58,8 @@ class Document(metaclass=_BackwardCompatible):
     :param blob: Binary data associated with the document, if the document has any binary data associated with it.
     :param meta: Additional custom metadata for the document. Must be JSON-serializable.
     :param score: Score of the document. Used for ranking, usually assigned by retrievers.
-    :param embedding: Vector representation of the document.
+    :param embedding: dense vector representation of the document.
+    :param sparse_embedding: sparse vector representation of the document.
     """
 
     id: str = field(default="")
@@ -67,6 +69,7 @@ class Document(metaclass=_BackwardCompatible):
     meta: Dict[str, Any] = field(default_factory=dict)
     score: Optional[float] = field(default=None)
     embedding: Optional[List[float]] = field(default=None)
+    sparse_embedding: Optional[SparseEmbedding] = field(default=None)
 
     def __repr__(self):
         fields = []
@@ -84,6 +87,8 @@ def __repr__(self):
             fields.append(f"score: {self.score}")
         if self.embedding is not None:
             fields.append(f"embedding: vector of size {len(self.embedding)}")
+        if self.sparse_embedding is not None:
+            fields.append(f"sparse_embedding: vector with {len(self.sparse_embedding.indices)} non-zero elements")
         fields_str = ", ".join(fields)
         return f"{self.__class__.__name__}(id={self.id}, {fields_str})"
 
@@ -114,7 +119,8 @@ def _create_id(self):
         mime_type = self.blob.mime_type if self.blob is not None else None
         meta = self.meta or {}
         embedding = self.embedding if self.embedding is not None else None
-        data = f"{text}{dataframe}{blob}{mime_type}{meta}{embedding}"
+        sparse_embedding = self.sparse_embedding.to_dict() if self.sparse_embedding is not None else ""
+        data = f"{text}{dataframe}{blob}{mime_type}{meta}{embedding}{sparse_embedding}"
         return hashlib.sha256(data.encode("utf-8")).hexdigest()
 
     def to_dict(self, flatten=True) -> Dict[str, Any]:
@@ -132,6 +138,9 @@ def to_dict(self, flatten=True) -> Dict[str, Any]:
         if (blob := data.get("blob")) is not None:
             data["blob"] = {"data": list(blob["data"]), "mime_type": blob["mime_type"]}
 
+        if (sparse_embedding := data.get("sparse_embedding")) is not None:
+            data["sparse_embedding"] = sparse_embedding.to_dict()
+
         if flatten:
             meta = data.pop("meta")
             return {**data, **meta}
@@ -149,6 +158,9 @@ def from_dict(cls, data: Dict[str, Any]) -> "Document":
             data["dataframe"] = read_json(io.StringIO(dataframe))
         if blob := data.get("blob"):
             data["blob"] = ByteStream(data=bytes(blob["data"]), mime_type=blob["mime_type"])
+        if sparse_embedding := data.get("sparse_embedding"):
+            data["sparse_embedding"] = SparseEmbedding.from_dict(sparse_embedding)
+
         # Store metadata for a moment while we try un-flattening allegedly flatten metadata.
         # We don't expect both a `meta=` keyword and flatten metadata keys so we'll raise a
         # ValueError later if this is the case.

haystack/dataclasses/sparse_embedding.py

@@ -0,0 +1,26 @@
+from typing import List
+
+
+class SparseEmbedding:
+    """
+    Class representing a sparse embedding.
+    """
+
+    def __init__(self, indices: List[int], values: List[float]):
+        """
+        :param indices: List of indices of non-zero elements in the embedding.
+        :param values: List of values of non-zero elements in the embedding.
+
+        :raises ValueError: If the indices and values lists are not of the same length.
+        """
+        if len(indices) != len(values):
+            raise ValueError("Length of indices and values must be the same.")
+        self.indices = indices
+        self.values = values
+
+    def to_dict(self):
+        return {"indices": self.indices, "values": self.values}
+
+    @classmethod
+    def from_dict(cls, sparse_embedding_dict):
+        return cls(indices=sparse_embedding_dict["indices"], values=sparse_embedding_dict["values"])

releasenotes/notes/sparse-embedding-fd55b670437492be.yaml

@@ -0,0 +1,7 @@
+---
+features:
+  - |
+    Introduce a new `SparseEmbedding` class which can be used to store a sparse
+    vector representation of a Document.
+    It will be instrumental to support Sparse Embedding Retrieval with
+    the subsequent introduction of Sparse Embedders and Sparse Embedding Retrievers.

test/dataclasses/test_document.py

@@ -3,6 +3,7 @@
 
 from haystack import Document
 from haystack.dataclasses.byte_stream import ByteStream
+from haystack.dataclasses.sparse_embedding import SparseEmbedding
 
 
 @pytest.mark.parametrize(
@@ -37,6 +38,7 @@ def test_init():
     assert doc.meta == {}
     assert doc.score == None
     assert doc.embedding == None
+    assert doc.sparse_embedding == None
 
 
 def test_init_with_wrong_parameters():
@@ -46,15 +48,17 @@ def test_init_with_wrong_parameters():
 
 def test_init_with_parameters():
     blob_data = b"some bytes"
+    sparse_embedding = SparseEmbedding(indices=[0, 2, 4], values=[0.1, 0.2, 0.3])
     doc = Document(
         content="test text",
         dataframe=pd.DataFrame([0]),
         blob=ByteStream(data=blob_data, mime_type="text/markdown"),
         meta={"text": "test text"},
         score=0.812,
         embedding=[0.1, 0.2, 0.3],
+        sparse_embedding=sparse_embedding,
     )
-    assert doc.id == "ec92455f3f4576d40031163c89b1b4210b34ea1426ee0ff68ebed86cb7ba13f8"
+    assert doc.id == "967b7bd4a21861ad9e863f638cefcbdd6bf6306bebdd30aa3fedf0c26bc636ed"
     assert doc.content == "test text"
     assert doc.dataframe is not None
     assert doc.dataframe.equals(pd.DataFrame([0]))
@@ -63,6 +67,7 @@ def test_init_with_parameters():
     assert doc.meta == {"text": "test text"}
     assert doc.score == 0.812
     assert doc.embedding == [0.1, 0.2, 0.3]
+    assert doc.sparse_embedding == sparse_embedding
 
 
 def test_init_with_legacy_fields():
@@ -76,6 +81,7 @@ def test_init_with_legacy_fields():
     assert doc.meta == {}
     assert doc.score == 0.812
     assert doc.embedding == [0.1, 0.2, 0.3]
+    assert doc.sparse_embedding == None
 
 
 def test_init_with_legacy_field():
@@ -93,6 +99,7 @@ def test_init_with_legacy_field():
     assert doc.meta == {"date": "10-10-2023", "type": "article"}
     assert doc.score == 0.812
     assert doc.embedding == [0.1, 0.2, 0.3]
+    assert doc.sparse_embedding == None
 
 
 def test_basic_equality_type_mismatch():
@@ -121,6 +128,7 @@ def test_to_dict():
         "blob": None,
         "score": None,
         "embedding": None,
+        "sparse_embedding": None,
     }
 
 
@@ -134,6 +142,7 @@ def test_to_dict_without_flattening():
         "meta": {},
         "score": None,
         "embedding": None,
+        "sparse_embedding": None,
     }
 
 
@@ -145,6 +154,7 @@ def test_to_dict_with_custom_parameters():
         meta={"some": "values", "test": 10},
         score=0.99,
         embedding=[10.0, 10.0],
+        sparse_embedding=SparseEmbedding(indices=[0, 2, 4], values=[0.1, 0.2, 0.3]),
     )
 
     assert doc.to_dict() == {
@@ -156,6 +166,7 @@ def test_to_dict_with_custom_parameters():
         "test": 10,
         "score": 0.99,
         "embedding": [10.0, 10.0],
+        "sparse_embedding": {"indices": [0, 2, 4], "values": [0.1, 0.2, 0.3]},
     }
 
 
@@ -167,6 +178,7 @@ def test_to_dict_with_custom_parameters_without_flattening():
         meta={"some": "values", "test": 10},
         score=0.99,
         embedding=[10.0, 10.0],
+        sparse_embedding=SparseEmbedding(indices=[0, 2, 4], values=[0.1, 0.2, 0.3]),
     )
 
     assert doc.to_dict(flatten=False) == {
@@ -177,6 +189,7 @@ def test_to_dict_with_custom_parameters_without_flattening():
         "meta": {"some": "values", "test": 10},
         "score": 0.99,
         "embedding": [10, 10],
+        "sparse_embedding": {"indices": [0, 2, 4], "values": [0.1, 0.2, 0.3]},
     }
 
 
@@ -194,6 +207,7 @@ def from_from_dict_with_parameters():
             "meta": {"text": "test text"},
             "score": 0.812,
             "embedding": [0.1, 0.2, 0.3],
+            "sparse_embedding": {"indices": [0, 2, 4], "values": [0.1, 0.2, 0.3]},
         }
     ) == Document(
         content="test text",
@@ -202,6 +216,7 @@ def from_from_dict_with_parameters():
         meta={"text": "test text"},
         score=0.812,
         embedding=[0.1, 0.2, 0.3],
+        sparse_embedding=SparseEmbedding(indices=[0, 2, 4], values=[0.1, 0.2, 0.3]),
     )
 
 
@@ -249,6 +264,7 @@ def test_from_dict_with_flat_meta():
             "blob": {"data": list(blob_data), "mime_type": "text/markdown"},
             "score": 0.812,
             "embedding": [0.1, 0.2, 0.3],
+            "sparse_embedding": {"indices": [0, 2, 4], "values": [0.1, 0.2, 0.3]},
             "date": "10-10-2023",
             "type": "article",
         }
@@ -258,6 +274,7 @@ def test_from_dict_with_flat_meta():
         blob=ByteStream(blob_data, mime_type="text/markdown"),
         score=0.812,
         embedding=[0.1, 0.2, 0.3],
+        sparse_embedding=SparseEmbedding(indices=[0, 2, 4], values=[0.1, 0.2, 0.3]),
         meta={"date": "10-10-2023", "type": "article"},
     )
 

test/dataclasses/test_sparse_embedding.py

@@ -0,0 +1,23 @@
+import pytest
+
+from haystack.dataclasses.sparse_embedding import SparseEmbedding
+
+
+class TestSparseEmbedding:
+    def test_init(self):
+        se = SparseEmbedding(indices=[0, 2, 4], values=[0.1, 0.2, 0.3])
+        assert se.indices == [0, 2, 4]
+        assert se.values == [0.1, 0.2, 0.3]
+
+    def test_init_with_wrong_parameters(self):
+        with pytest.raises(ValueError):
+            SparseEmbedding(indices=[0, 2], values=[0.1, 0.2, 0.3, 0.4])
+
+    def test_to_dict(self):
+        se = SparseEmbedding(indices=[0, 2, 4], values=[0.1, 0.2, 0.3])
+        assert se.to_dict() == {"indices": [0, 2, 4], "values": [0.1, 0.2, 0.3]}
+
+    def test_from_dict(self):
+        se = SparseEmbedding.from_dict({"indices": [0, 2, 4], "values": [0.1, 0.2, 0.3]})
+        assert se.indices == [0, 2, 4]
+        assert se.values == [0.1, 0.2, 0.3]

test/tracing/test_utils.py

@@ -25,15 +25,15 @@ class TestTypeCoercion:
             (NonSerializableClass(), "NonSerializableClass"),
             (
                 Document(id="1", content="text"),
-                '{"id": "1", "content": "text", "dataframe": null, "blob": null, "score": null, "embedding": null}',
+                '{"id": "1", "content": "text", "dataframe": null, "blob": null, "score": null, "embedding": null, "sparse_embedding": null}',
             ),
             (
                 [Document(id="1", content="text")],
-                '[{"id": "1", "content": "text", "dataframe": null, "blob": null, "score": null, "embedding": null}]',
+                '[{"id": "1", "content": "text", "dataframe": null, "blob": null, "score": null, "embedding": null, "sparse_embedding": null}]',
             ),
             (
                 {"key": Document(id="1", content="text")},
-                '{"key": {"id": "1", "content": "text", "dataframe": null, "blob": null, "score": null, "embedding": null}}',
+                '{"key": {"id": "1", "content": "text", "dataframe": null, "blob": null, "score": null, "embedding": null, "sparse_embedding": null}}',
             ),
         ],
     )