Add document for PyG remote backend

tools/python_api/src_py/database.py

@@ -15,7 +15,7 @@ class Database:
         Set the logging level.
 
     get_torch_geometric_remote_backend(num_threads)
-        Get the torch_geometric remote backend.
+        Use the database as the remote backend for torch_geometric. 
 
     """
 
@@ -81,10 +81,33 @@ def set_logging_level(self, level):
         self._database.set_logging_level(level)
 
     def get_torch_geometric_remote_backend(self, num_threads=None):
-        from .torch_geometric_feature_store import KuzuFeatureStore
-        from .torch_geometric_graph_store import KuzuGraphStore
         """
-        Get the torch_geometric remote backend.
+        Use the database as the remote backend for torch_geometric. 
+
+        For the interface of the remote backend, please refer to 
+        https://pytorch-geometric.readthedocs.io/en/latest/advanced/remote.html.
+        The current implementation is read-only and does not support edge
+        features. The IDs of the nodes are based on the internal IDs (i.e., node
+        offsets). For the remote node IDs to be consistent with the positions in
+        the output tensors, please ensure that no deletion has been performed
+        on the node tables. 
+
+        The remote backend can also be plugged into the data loader of 
+        torch_geometric, which is useful for mini-batch training. For example:
+
+        .. code-block:: python
+
+            loader_kuzu = NeighborLoader(
+                data=(feature_store, graph_store),
+                num_neighbors={('paper', 'cites', 'paper'): [12, 12, 12]},
+                batch_size=LOADER_BATCH_SIZE,
+                input_nodes=('paper', input_nodes),
+                num_workers=4,
+                filter_per_worker=False,
+            )
+
+        Please note that the database instance is not fork-safe, so if more than
+        one worker is used, `filter_per_worker` must be set to False.
 
         Parameters
         ----------
@@ -99,6 +122,8 @@ def get_torch_geometric_remote_backend(self, num_threads=None):
         graph_store : KuzuGraphStore
             Graph store compatible with torch_geometric.
         """
+        from .torch_geometric_feature_store import KuzuFeatureStore
+        from .torch_geometric_graph_store import KuzuGraphStore
         return KuzuFeatureStore(self, num_threads), KuzuGraphStore(self, num_threads)
 
     def _scan_node_table(self, table_name, prop_name, prop_type, dim, indices, num_threads):