Merge main into branch

CHANGELOG.md

@@ -16,6 +16,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   [\#19](https://github.com/mllam/weather-model-graphs/pull/19)
   @joeloskarsson
 
+- `save.to_pyg` can now handle any number of 1D or 2D edge or node features when
+  converting pytorch-geometric `Data` objects to `torch.Tensor` objects.
+  [\#31](https://github.com/mllam/weather-model-graphs/pull/31)
+  @maxiimilian
+
 ### Changed
 
 - Fix wrong number of mesh levels when grid is multiple of refinement factor

src/weather_model_graphs/save.py

@@ -1,5 +1,6 @@
 import pickle
 from pathlib import Path
+from typing import List
 
 import networkx
 from loguru import logger
@@ -8,6 +9,7 @@
 
 try:
     import torch
+    import torch_geometric as pyg
     import torch_geometric.utils.convert as pyg_convert
 
     HAS_PYG = True
@@ -19,8 +21,8 @@ def to_pyg(
     graph: networkx.DiGraph,
     output_directory: str,
     name: str,
-    edge_features=["vdiff"],
-    node_features=["pos"],
+    edge_features: List[str] | None = None,
+    node_features: List[str] | None = None,
     list_from_attribute=None,
 ):
     """
@@ -59,6 +61,13 @@ def to_pyg(
             "install weather-mode-graphs[pytorch] to enable writing to torch files"
         )
 
+    # Default values for arguments
+    if edge_features is None:
+        edge_features = ["len", "vdiff"]
+
+    if node_features is None:
+        node_features = ["pos"]
+
     # check that the node labels are integers and unique so that they can be used as indices
     if not all(isinstance(node, int) for node in graph.nodes):
         node_types = set([type(node) for node in graph.nodes])
@@ -77,16 +86,22 @@ def to_pyg(
     def _get_edge_indecies(pyg_g):
         return pyg_g.edge_index
 
-    def _get_edge_features(pyg_g):
-        if edge_features != ["vdiff"]:
-            raise NotImplementedError(edge_features_values)
-        # TODO: handle features of different types more generally, i.e. both single ("len") values and tuples (like "vdiff")
-        return torch.cat((pyg_g.len.unsqueeze(1), pyg_g.vdiff), dim=1).to(torch.float32)
-
-    def _get_node_features(pyg_g):
-        if node_features != ["pos"]:
-            raise NotImplementedError(node_features_values)
-        return pyg_g.pos.to(torch.float32)
+    def _concat_pyg_features(
+        pyg_g: "pyg.data.Data", features: List[str]
+    ) -> torch.Tensor:
+        """Convert features from pyg.Data object to torch.Tensor.
+        Each feature should be column in the resulting 2D tensor (n_edges or n_nodes, n_features).
+        Note, this function can handle node AND edge features.
+        """
+        v_concat = []
+        for f in features:
+            v = pyg_g[f]
+            # Convert 1D features into 1xN tensor
+            if v.ndim == 1:
+                v = v.unsqueeze(1)
+            v_concat.append(v)
+
+        return torch.cat(v_concat, dim=1).to(torch.float32)
 
     if list_from_attribute is not None:
         # create a list of graph objects by splitting the graph by the list_from_attribute
@@ -104,9 +119,13 @@ def _get_node_features(pyg_g):
     else:
         pyg_graphs = [pyg_convert.from_networkx(graph)]
 
-    edge_features_values = [_get_edge_features(pyg_g) for pyg_g in pyg_graphs]
+    edge_features_values = [
+        _concat_pyg_features(pyg_g, features=edge_features) for pyg_g in pyg_graphs
+    ]
     edge_indecies = [_get_edge_indecies(pyg_g) for pyg_g in pyg_graphs]
-    node_features_values = [_get_node_features(pyg_g) for pyg_g in pyg_graphs]
+    node_features_values = [
+        _concat_pyg_features(pyg_g, features=node_features) for pyg_g in pyg_graphs
+    ]
 
     if list_from_attribute is None:
         edge_features_values = edge_features_values[0]