SB server update

.gitignore

@@ -1,8 +1,9 @@
 .idea/*
 logs/*
 data/*
-gifflar/.ipynb_checkpoints/*
+**/.ipynb_checkpoints/
 *.ipynb
 *.tsv
 tests/data/*
-tests/logs/*
+tests/logs/*
+**/__pycache__/

collect.py

@@ -65,7 +65,11 @@ def order_models(name: str) -> tuple:
         worksheet = writer.sheets[metric]
         bold_format = workbook.add_format({"bold": True})
         for r, (_, row) in enumerate(df.iterrows()):
-            max_val = row.values.max()
-            for c, val in enumerate(row.values):
-                if val == max_val:
-                    worksheet.write(r + 1, c + 1, val, bold_format)
+            try:
+                max_val = row.values.max()
+                for c, val in enumerate(row.values):
+                    if val == max_val:
+                        worksheet.write(r + 1, c + 1, val, bold_format)
+            except:
+                pass
+

gifflar/data/hetero.py

@@ -129,6 +129,7 @@ def hetero_collate(data: Optional[Union[list[list[HeteroData]], list[HeteroData]
 
     # For each baseline, collate its node features and edge indices as well
     for b in baselines:
+        break
         kwargs[f"{b}_x"] = torch.cat([d[f"{b}_x"] for d in data], dim=0)
         edges = []
         batch = []

gifflar/pretransforms.py

@@ -170,13 +170,10 @@ def __call__(self, data: HeteroData) -> HeteroData:
         ])
         data["rgcn_node_type"] = ["atoms"] * len(data["atoms"].x) + ["bonds"] * len(data["bonds"].x) + ["monosacchs"] * len(data["monosacchs"].x)
         data["rgcn_num_nodes"] = len(data["rgcn_x"])
-        data["rgcn_edge_type"] = torch.tensor(
-            [0] * data["atoms", "coboundary", "atoms"].edge_index.shape[1]
-            + [1] * data["atoms", "to", "bonds"].edge_index.shape[1]
-            + [2] * data["bonds", "boundary", "bonds"].edge_index.shape[1]
-            + [3] * data["bonds", "to", "monosacchs"].edge_index.shape[1]
-            + [4] * data["monosacchs", "boundary", "monosacchs"].edge_index.shape[1]
-        )
+        tmp = [0] * data["atoms", "coboundary", "atoms"].edge_index.shape[1] + [1] * data["atoms", "to", "bonds"].edge_index.shape[1] + [2] * data["bonds", "boundary", "bonds"].edge_index.shape[1] + [3] * data["bonds", "to", "monosacchs"].edge_index.shape[1]
+        if len(data["monosacchs"].x) > 1:
+            tmp += [4] * data["monosacchs", "boundary", "monosacchs"].edge_index.shape[1]
+        data["rgcn_edge_type"] = torch.tensor(tmp)
         tmp = []
         offset = {"atoms": 0,
                   "bonds": data["atoms"]["num_nodes"],
@@ -186,10 +183,11 @@ def __call__(self, data: HeteroData) -> HeteroData:
                     ("bonds", "to", "monosacchs"),
                     ("bonds", "boundary", "bonds"),
                     ("monosacchs", "boundary", "monosacchs")]:
-            tmp.append(torch.stack([
-                data[key].edge_index[0] + offset[key[0]],
-                data[key].edge_index[1] + offset[key[2]],
-            ]))
+            if len(data[key].edge_index.shape) == 2:
+                tmp.append(torch.stack([
+                    data[key].edge_index[0] + offset[key[0]],
+                    data[key].edge_index[1] + offset[key[2]],
+                ]))
         data["rgcn_edge_index"] = torch.cat(tmp, dim=1)
         return data
 

gifflar/train.py

@@ -146,7 +146,7 @@ def train(**kwargs: Any) -> None:
         ],
         max_epochs=kwargs["model"]["epochs"],
         logger=logger,
-        accelerator="cpu",
+        # accelerator="cpu",
     )
     start = time.time()
     trainer.fit(model, datamodule)