Allow deleting vectors with prefix or filter (#43)

* Allow deleting vectors with prefix or filter

We have started supporting deleting vectors with id prefix or
metadata filter.

* fix type hints

Author: mdumandag

README.md

@@ -404,9 +404,8 @@ res = index.delete(
     ids=["id1", "id2"],
 )
 
-print(
-    res.deleted,  # How many vectors are deleted out of the given ids.
-)
+# How many vectors are deleted out of the given ids.
+print(res.deleted)
 ```
 
 or, for singular deletion:
@@ -416,7 +415,23 @@ res = index.delete(
     "id1",
 )
 
-print(res)  # A boolean indicating whether the vector is deleted or not.
+# 1 if the vector is deleted, 0 otherwise.
+print(res.deleted)
+```
+
+Apart from the vector ids, vectors can also be deleted with an id prefix
+or metadata filter.
+
+```python
+# Delete all the vectors whose id starts with `id-0`
+index.delete(
+    prefix="id-0",
+)
+
+# Delete all the vectors whose metadata matches with the filter
+index.delete(
+    filter="salary < 3000",
+)
 ```
 
 Also, a namespace can be specified to delete from.

tests/core/test_delete.py

@@ -137,3 +137,125 @@ async def test_delete_async(async_index: AsyncIndex, ns: str):
     assert res[0] is None
     assert res[1] is None
     assert res[2] is not None
+
+
+@pytest.mark.parametrize("ns", NAMESPACES)
+def test_delete_prefix(index: Index, ns: str):
+    index.upsert(
+        vectors=[
+            ("id-00", [0.1, 0.2]),
+            ("id-01", [0.1, 0.3]),
+            ("id-10", [0.1, 0.4]),
+            ("id-11", [0.1, 0.5]),
+        ],
+        namespace=ns,
+    )
+
+    result = index.delete(
+        prefix="id-0",
+        namespace=ns,
+    )
+
+    assert result.deleted == 2
+
+    vectors = index.fetch(
+        ids=["id-00", "id-01", "id-10", "id-11"],
+        namespace=ns,
+    )
+    assert len(vectors) == 4
+    assert vectors[0] is None
+    assert vectors[1] is None
+    assert vectors[2] is not None
+    assert vectors[3] is not None
+
+
+@pytest.mark.parametrize("ns", NAMESPACES)
+def test_delete_filter(index: Index, ns: str):
+    index.upsert(
+        vectors=[
+            ("id-00", [0.1, 0.2], {"meta": 0}),
+            ("id-01", [0.1, 0.3], {"meta": 1}),
+            ("id-10", [0.1, 0.4], {"meta": 2}),
+            ("id-11", [0.1, 0.5], {"meta": 3}),
+        ],
+        namespace=ns,
+    )
+
+    result = index.delete(
+        filter="meta >= 2",
+        namespace=ns,
+    )
+
+    assert result.deleted == 2
+
+    vectors = index.fetch(
+        ids=["id-00", "id-01", "id-10", "id-11"],
+        namespace=ns,
+    )
+    assert len(vectors) == 4
+    assert vectors[0] is not None
+    assert vectors[1] is not None
+    assert vectors[2] is None
+    assert vectors[3] is None
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("ns", NAMESPACES)
+async def test_delete_prefix_async(async_index: AsyncIndex, ns: str):
+    await async_index.upsert(
+        vectors=[
+            ("id-00", [0.1, 0.2]),
+            ("id-01", [0.1, 0.3]),
+            ("id-10", [0.1, 0.4]),
+            ("id-11", [0.1, 0.5]),
+        ],
+        namespace=ns,
+    )
+
+    result = await async_index.delete(
+        prefix="id-0",
+        namespace=ns,
+    )
+
+    assert result.deleted == 2
+
+    vectors = await async_index.fetch(
+        ids=["id-00", "id-01", "id-10", "id-11"],
+        namespace=ns,
+    )
+    assert len(vectors) == 4
+    assert vectors[0] is None
+    assert vectors[1] is None
+    assert vectors[2] is not None
+    assert vectors[3] is not None
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("ns", NAMESPACES)
+async def test_delete_filter_async(async_index: AsyncIndex, ns: str):
+    await async_index.upsert(
+        vectors=[
+            ("id-00", [0.1, 0.2], {"meta": 0}),
+            ("id-01", [0.1, 0.3], {"meta": 1}),
+            ("id-10", [0.1, 0.4], {"meta": 2}),
+            ("id-11", [0.1, 0.5], {"meta": 3}),
+        ],
+        namespace=ns,
+    )
+
+    result = await async_index.delete(
+        filter="meta >= 2",
+        namespace=ns,
+    )
+
+    assert result.deleted == 2
+
+    vectors = await async_index.fetch(
+        ids=["id-00", "id-01", "id-10", "id-11"],
+        namespace=ns,
+    )
+    assert len(vectors) == 4
+    assert vectors[0] is not None
+    assert vectors[1] is not None
+    assert vectors[2] is None
+    assert vectors[3] is None

upstash_vector/core/index_operations.py

@@ -415,8 +415,10 @@ def resumable_query(
 
     def delete(
         self,
-        ids: Union[str, List[str]],
+        ids: Optional[Union[str, List[str]]] = None,
         namespace: str = DEFAULT_NAMESPACE,
+        prefix: Optional[str] = None,
+        filter: Optional[str] = None,
     ) -> DeleteResult:
         """
         Deletes the given vector(s) with given ids.
@@ -425,6 +427,8 @@ def delete(
 
         :param ids: Singular or list of ids of vector(s) to be deleted.
         :param namespace: The namespace to use. When not specified, the default namespace is used.
+        :param prefix: Prefix of vector ids to delete.
+        :param filter: Metadata filter for the vectors to delete.
 
         Example usage:
 
@@ -436,11 +440,23 @@ def delete(
         index.delete("0")
         ```
         """
-        if not isinstance(ids, list):
-            ids = [ids]
+        payload: Dict[str, Any] = {}
+        if ids is not None:
+            if not isinstance(ids, list):
+                ids = [ids]
+
+            payload["ids"] = ids
+
+        if prefix is not None:
+            payload["prefix"] = prefix
+
+        if filter is not None:
+            payload["filter"] = filter
 
         return DeleteResult._from_json(
-            self._execute_request(payload=ids, path=_path_for(namespace, DELETE_PATH))
+            self._execute_request(
+                payload=payload, path=_path_for(namespace, DELETE_PATH)
+            )
         )
 
     def reset(self, namespace: str = DEFAULT_NAMESPACE, all: bool = False) -> str:
@@ -1000,8 +1016,10 @@ async def resumable_query(
 
     async def delete(
         self,
-        ids: Union[str, List[str]],
+        ids: Optional[Union[str, List[str]]] = None,
         namespace: str = DEFAULT_NAMESPACE,
+        prefix: Optional[str] = None,
+        filter: Optional[str] = None,
     ) -> DeleteResult:
         """
         Deletes the given vector(s) with given ids asynchronously.
@@ -1010,6 +1028,8 @@ async def delete(
 
         :param ids: Singular or list of ids of vector(s) to be deleted.
         :param namespace: The namespace to use. When not specified, the default namespace is used.
+        :param prefix: Prefix of vector ids to delete.
+        :param filter: Metadata filter for the vectors to delete.
 
         Example usage:
 
@@ -1021,12 +1041,22 @@ async def delete(
         await index.delete("0")
         ```
         """
-        if not isinstance(ids, list):
-            ids = [ids]
+        payload: Dict[str, Any] = {}
+        if ids is not None:
+            if not isinstance(ids, list):
+                ids = [ids]
+
+            payload["ids"] = ids
+
+        if prefix is not None:
+            payload["prefix"] = prefix
+
+        if filter is not None:
+            payload["filter"] = filter
 
         return DeleteResult._from_json(
             await self._execute_request_async(
-                payload=ids, path=_path_for(namespace, DELETE_PATH)
+                payload=payload, path=_path_for(namespace, DELETE_PATH)
             )
         )