Support edges to/from nested fields

data/animals.jsonl

@@ -1,9 +1,9 @@
 {"id": "aardvark", "text": "the aardvark is a nocturnal mammal known for its burrowing habits and long snout used to sniff out ants.", "metadata": {"type": "mammal", "number_of_legs": 4, "keywords": ["burrowing", "nocturnal", "ants", "savanna"], "habitat": "savanna"}}
 {"id": "albatross", "text": "the albatross is a large seabird with the longest wingspan of any bird, allowing it to glide effortlessly over oceans.", "metadata": {"type": "bird", "number_of_legs": 2, "keywords": ["seabird", "wingspan", "ocean"], "habitat": "marine"}}
-{"id": "alligator", "text": "alligators are large reptiles with powerful jaws and are commonly found in freshwater wetlands.", "metadata": {"type": "reptile", "number_of_legs": 4, "keywords": ["reptile", "jaws", "wetlands"], "diet": "carnivorous"}}
-{"id": "alpaca", "text": "alpacas are domesticated mammals valued for their soft wool and friendly demeanor.", "metadata": {"type": "mammal", "number_of_legs": 4, "keywords": ["wool", "domesticated", "friendly"], "origin": "south america"}}
-{"id": "ant", "text": "ants are social insects that live in colonies and are known for their teamwork and strength.", "metadata": {"type": "insect", "number_of_legs": 6, "keywords": ["social", "colonies", "strength", "pollinator"], "diet": "omnivorous"}}
-{"id": "anteater", "text": "anteaters use their long tongues to eat thousands of ants and termites each day.", "metadata": {"type": "mammal", "number_of_legs": 4, "keywords": ["ants", "tongue", "termites"], "diet": "insectivore"}}
+{"id": "alligator", "text": "alligators are large reptiles with powerful jaws and are commonly found in freshwater wetlands.", "metadata": {"type": "reptile", "number_of_legs": 4, "keywords": ["reptile", "jaws", "wetlands"], "diet": "carnivorous", "nested": { "a": 5 }}}
+{"id": "alpaca", "text": "alpacas are domesticated mammals valued for their soft wool and friendly demeanor.", "metadata": {"type": "mammal", "number_of_legs": 4, "keywords": ["wool", "domesticated", "friendly"], "origin": "south america", "nested": { "a": 5 }}}
+{"id": "ant", "text": "ants are social insects that live in colonies and are known for their teamwork and strength.", "metadata": {"type": "insect", "number_of_legs": 6, "keywords": ["social", "colonies", "strength", "pollinator"], "diet": "omnivorous", "nested": { "a": 6 }}}
+{"id": "anteater", "text": "anteaters use their long tongues to eat thousands of ants and termites each day.", "metadata": {"type": "mammal", "number_of_legs": 4, "keywords": ["ants", "tongue", "termites"], "diet": "insectivore", "nested": { "b": 5 }}}
 {"id": "antelope", "text": "antelopes are graceful herbivorous mammals that are often prey for large predators in the wild.", "metadata": {"type": "mammal", "number_of_legs": 4, "keywords": ["graceful", "herbivore", "prey"], "habitat": "grasslands"}}
 {"id": "armadillo", "text": "armadillos have hard, protective shells and are known for their ability to roll into a ball.", "metadata": {"type": "mammal", "number_of_legs": 4, "keywords": ["protective", "shell", "rolling"], "diet": "insectivore"}}
 {"id": "baboon", "text": "baboons are highly social primates with complex group dynamics and strong bonds.", "metadata": {"type": "mammal", "number_of_legs": 4, "keywords": ["social", "primates", "group"], "diet": "omnivorous"}}

packages/graph-retriever/src/graph_retriever/adapters/in_memory.py

@@ -113,7 +113,11 @@ def _matches(self, filter: dict[str, Any] | None, content: Content) -> bool:
             return True
 
         for key, filter_value in filter.items():
-            content_value = content.metadata.get(key, SENTINEL)
+            content_value = content.metadata
+            for key_part in key.split("."):
+                content_value = content_value.get(key_part, SENTINEL)
+                if content_value is SENTINEL:
+                    break
             if not self._value_matches(filter_value, content_value):
                 return False
         return True

packages/graph-retriever/src/graph_retriever/edges/metadata.py

@@ -41,6 +41,15 @@ class Id:
 """
 
 
+def _nested_get(metadata: dict[str, Any], key: str) -> Any:
+    value = metadata
+    for key_part in key.split("."):
+        value = value.get(key_part, SENTINEL)
+        if value is SENTINEL:
+            break
+    return value
+
+
 class MetadataEdgeFunction:
     """
     Helper for extracting and encoding edges in metadata.
@@ -116,7 +125,7 @@ def mk_edge(v) -> Edge:
             if isinstance(source_key, Id):
                 edges.add(mk_edge(id))
             else:
-                value = metadata.get(source_key, SENTINEL)
+                value = _nested_get(metadata, source_key)
                 if isinstance(value, BASIC_TYPES):
                     edges.add(mk_edge(value))
                 elif isinstance(value, Iterable):

packages/graph-retriever/src/graph_retriever/testing/adapter_tests.py

@@ -40,7 +40,7 @@ def assert_ids_any_order(
     assert set(result_ids) == set(expected), "should contain exactly expected IDs"
 
 
-@dataclass
+@dataclass(kw_only=True)
 class AdapterComplianceCase(abc.ABC):
     """
     Base dataclass for test cases.
@@ -56,6 +56,7 @@ class AdapterComplianceCase(abc.ABC):
 
     id: str
     expected: list[str]
+    requires_nested: bool = False
 
 
 @dataclass
@@ -261,6 +262,46 @@ class AdjacentCase(AdapterComplianceCase):
             "komodo dragon",  # reptile
         ],
     ),
+    AdjacentCase(
+        id="nested",
+        query="domesticated hunters",
+        edges={
+            MetadataEdge("nested.a", 5),
+        },
+        expected=[
+            "alligator",
+            "alpaca",
+        ],
+        requires_nested=True,
+    ),
+    AdjacentCase(
+        id="nested_same_field",
+        query="domesticated hunters",
+        edges={
+            MetadataEdge("nested.a", 5),
+            MetadataEdge("nested.a", 6),
+        },
+        expected=[
+            "alligator",
+            "alpaca",
+            "ant",
+        ],
+        requires_nested=True,
+    ),
+    AdjacentCase(
+        id="nested_diff_field",
+        query="domesticated hunters",
+        edges={
+            MetadataEdge("nested.a", 5),
+            MetadataEdge("nested.b", 5),
+        },
+        expected=[
+            "alligator",
+            "alpaca",
+            "anteater",
+        ],
+        requires_nested=True,
+    ),
 ]
 
 
@@ -273,6 +314,10 @@ class AdapterComplianceSuite(abc.ABC):
     loaded.
     """
 
+    def supports_nested_metadata(self) -> bool:
+        """Return whether nested metadata is expected to work."""
+        return True
+
     def expected(self, method: str, case: AdapterComplianceCase) -> list[str]:
         """
         Override to change the expected behavior of a case.
@@ -299,6 +344,8 @@ def expected(self, method: str, case: AdapterComplianceCase) -> list[str]:
         :
             The expected animals.
         """
+        if not self.supports_nested_metadata() and case.requires_nested:
+            pytest.xfail("nested metadata not supported")
         return case.expected
 
     @pytest.fixture(params=GET_CASES, ids=lambda c: c.id)

packages/graph-retriever/tests/edges/test_metadata.py

@@ -38,6 +38,14 @@ def test_edge_function():
     )
 
 
+def test_nested_edge():
+    edge_function = MetadataEdgeFunction([("a.b", "b.c")])
+    assert edge_function(mk_node({"a": {"b": 5}, "b": {"c": 7}})) == Edges(
+        {MetadataEdge("b.c", 7)},
+        {MetadataEdge("b.c", 5)},
+    )
+
+
 def test_link_to_id():
     edge_function = MetadataEdgeFunction([("mentions", Id())])
     result = edge_function(mk_node({"mentions": ["a", "c"]}))

packages/langchain-graph-retriever/src/langchain_graph_retriever/adapters/in_memory.py

@@ -99,7 +99,11 @@ def _equals_or_contains(
             True if and only if `metadata[key] == value` or `metadata[key]` is a
             list containing `value`.
         """
-        actual = metadata.get(key, SENTINEL)
+        actual = metadata
+        for key_part in key.split("."):
+            actual = actual.get(key_part, SENTINEL)
+            if actual is SENTINEL:
+                break
         if actual == value:
             return True
 

packages/langchain-graph-retriever/tests/adapters/test_cassandra.py

@@ -48,6 +48,9 @@ def cluster(
 
 
 class TestCassandraAdapter(AdapterComplianceSuite):
+    def supports_nested_metadata(self) -> bool:
+        return False
+
     @pytest.fixture(scope="class")
     def adapter(
         self,

packages/langchain-graph-retriever/tests/adapters/test_chroma.py

@@ -9,6 +9,9 @@
 
 
 class TestChroma(AdapterComplianceSuite):
+    def supports_nested_metadata(self) -> bool:
+        return False
+
     @pytest.fixture(scope="class")
     def adapter(
         self,
@@ -25,6 +28,16 @@ def adapter(
         )
 
         shredder = ShreddingTransformer()
+
+        # Chroma doesn't even support *writing* nested data currently, so we
+        # filter it out.
+        def remove_nested_metadata(doc: Document) -> Document:
+            metadata = doc.metadata.copy()
+            metadata.pop("nested", None)
+            return Document(id=doc.id, page_content=doc.page_content, metadata=metadata)
+
+        animal_docs = [remove_nested_metadata(doc) for doc in animal_docs]
+
         docs = list(shredder.transform_documents(animal_docs))
         store = Chroma.from_documents(
             docs,

packages/langchain-graph-retriever/tests/adapters/test_open_search.py

@@ -8,6 +8,9 @@
 
 
 class TestOpenSearch(AdapterComplianceSuite):
+    def supports_nested_metadata(self) -> bool:
+        return False
+
     @pytest.fixture(scope="class")
     def adapter(
         self,