Add del vector pgvector + adding modification time to confluence and google drive docs (#9604)

Description:
- adding implementation of delete for pgvector
- adding modification time in docs metadata for confluence and google
drive.

Issue:
https://github.com/langchain-ai/langchain/issues/9312

Tag maintainer: @baskaryan, @eyurtsev, @hwchase17, @rlancemartin.

---------

Co-authored-by: Eugene Yurtsev <eyurtsev@gmail.com>

libs/langchain/langchain/document_loaders/confluence.py

@@ -338,7 +338,9 @@ class ConfluenceLoader(BaseLoader):
                     ),
                     before_sleep=before_sleep_log(logger, logging.WARNING),
                 )(self.confluence.get_page_by_id)
-                page = get_page(page_id=page_id, expand=content_format.value)
+                page = get_page(
+                    page_id=page_id, expand=f"{content_format.value},version"
+                )
                 if not include_restricted_content and not self.is_public_page(page):
                     continue
                 doc = self.process_page(
@@ -505,13 +507,18 @@ class ConfluenceLoader(BaseLoader):
             ]
             text = text + "".join(comment_texts)
 
+        metadata = {
+            "title": page["title"],
+            "id": page["id"],
+            "source": self.base_url.strip("/") + page["_links"]["webui"],
+        }
+
+        if "version" in page and "when" in page["version"]:
+            metadata["when"] = page["version"]["when"]
+
         return Document(
             page_content=text,
-            metadata={
-                "title": page["title"],
-                "id": page["id"],
-                "source": self.base_url.strip("/") + page["_links"]["webui"],
-            },
+            metadata=metadata,
         )
 
     def process_attachment(

libs/langchain/langchain/document_loaders/googledrive.py

@@ -200,7 +200,11 @@ class GoogleDriveLoader(BaseLoader, BaseModel):
         creds = self._load_credentials()
         service = build("drive", "v3", credentials=creds)
 
-        file = service.files().get(fileId=id, supportsAllDrives=True).execute()
+        file = (
+            service.files()
+            .get(fileId=id, supportsAllDrives=True, fields="modifiedTime,name")
+            .execute()
+        )
         request = service.files().export_media(fileId=id, mimeType="text/plain")
         fh = BytesIO()
         downloader = MediaIoBaseDownload(fh, request)
@@ -219,6 +223,7 @@ class GoogleDriveLoader(BaseLoader, BaseModel):
         metadata = {
             "source": f"https://docs.google.com/document/d/{id}/edit",
             "title": f"{file.get('name')}",
+            "when": f"{file.get('modifiedTime')}",
         }
         return Document(page_content=text, metadata=metadata)
 

libs/langchain/langchain/vectorstores/pgvector.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import contextlib
 import enum
 import logging
 import uuid
@@ -8,6 +9,7 @@ from typing import (
     Any,
     Callable,
     Dict,
+    Generator,
     Iterable,
     List,
     Optional,
@@ -16,6 +18,7 @@ from typing import (
 )
 
 import sqlalchemy
+from sqlalchemy import delete
 from sqlalchemy.dialects.postgresql import UUID
 from sqlalchemy.orm import Session, declarative_base
 
@@ -170,6 +173,33 @@ class PGVector(VectorStore):
             session.delete(collection)
             session.commit()
 
+    @contextlib.contextmanager
+    def _make_session(self) -> Generator[Session, None, None]:
+        """Create a context manager for the session, bind to _conn string."""
+        yield Session(self._conn)
+
+    def delete(
+        self,
+        ids: Optional[List[str]] = None,
+        **kwargs: Any,
+    ) -> None:
+        """Delete vectors by ids or uuids.
+
+        Args:
+            ids: List of ids to delete.
+        """
+        with Session(self._conn) as session:
+            if ids is not None:
+                self.logger.debug(
+                    "Trying to delete vectors by ids (represented by the model "
+                    "using the custom ids field)"
+                )
+                stmt = delete(self.EmbeddingStore).where(
+                    self.EmbeddingStore.custom_id.in_(ids)
+                )
+                session.execute(stmt)
+            session.commit()
+
     def get_collection(self, session: Session) -> Optional["CollectionStore"]:
         return self.CollectionStore.get_by_name(session, self.collection_name)
 

libs/langchain/tests/integration_tests/vectorstores/test_pgvector.py

@@ -186,6 +186,34 @@ def test_pgvector_with_filter_in_set() -> None:
     ]
 
 
+def test_pgvector_delete_docs() -> None:
+    """Add and delete documents."""
+    texts = ["foo", "bar", "baz"]
+    metadatas = [{"page": str(i)} for i in range(len(texts))]
+    docsearch = PGVector.from_texts(
+        texts=texts,
+        collection_name="test_collection_filter",
+        embedding=FakeEmbeddingsWithAdaDimension(),
+        metadatas=metadatas,
+        ids=["1", "2", "3"],
+        connection_string=CONNECTION_STRING,
+        pre_delete_collection=True,
+    )
+    docsearch.delete(["1", "2"])
+    with docsearch._make_session() as session:
+        records = list(session.query(docsearch.EmbeddingStore).all())
+        # ignoring type error since mypy cannot determine whether
+        # the list is sortable
+        assert sorted(record.custom_id for record in records) == ["3"]  # type: ignore
+
+    docsearch.delete(["2", "3"])  # Should not raise on missing ids
+    with docsearch._make_session() as session:
+        records = list(session.query(docsearch.EmbeddingStore).all())
+        # ignoring type error since mypy cannot determine whether
+        # the list is sortable
+        assert sorted(record.custom_id for record in records) == []  # type: ignore
+
+
 def test_pgvector_relevance_score() -> None:
     """Test to make sure the relevance score is scaled to 0-1."""
     texts = ["foo", "bar", "baz"]