code-transformer

code-transformer copied to clipboard

When run on pytorch '1.10.2+cu102' I get following error:

Traceback (most recent call last):
  File "/code_transformer/preprocessing/graph/transform.py", line 30, in __call__
    distance_matrix = distance_metric(adj)
  File "/code_transformer/preprocessing/graph/distances.py", line 76, in __call__
    sp_length = all_pairs_shortest_paths(G=G)
  File "/code_transformer/preprocessing/graph/alg.py", line 45, in all_pairs_shortest_paths
    values = torch.tensor([(dct[0], key, value) for dct in sps for key, value in dct[1].items()],
TypeError: 'float' object cannot be interpreted as an integer

Following diff seems to solve the issue. Somehow pytorch now is not happy with converting floats to integers :)

Index: code_transformer/preprocessing/graph/alg.py
IDEA additional info:
Subsystem: com.intellij.openapi.diff.impl.patch.CharsetEP
<+>UTF-8
===================================================================
diff --git a/code_transformer/preprocessing/graph/alg.py b/code_transformer/preprocessing/graph/alg.py
--- a/code_transformer/preprocessing/graph/alg.py	(revision 362ec5300e94c6901566b38e10cb3b93440e7c52)
+++ b/code_transformer/preprocessing/graph/alg.py	(date 1663341487624)
@@ -42,7 +42,7 @@
             create_using = nx.Graph
         G = nx.from_edgelist(edges, create_using=create_using)
     sps = nx.all_pairs_dijkstra_path_length(G, cutoff=cutoff)
-    values = torch.tensor([(dct[0], key, value) for dct in sps for key, value in dct[1].items()],
+    values = torch.tensor([(dct[0], key, int(value)) for dct in sps for key, value in dct[1].items()],
                           dtype=torch.long)
     return values
 
@@ -95,7 +95,7 @@
     sibling_edges = next_sibling_edges(tree_edges).numpy()
     G_siblings = nx.from_edgelist(sibling_edges, create_using=nx.DiGraph)
     sps = list(nx.all_pairs_dijkstra_path_length(G_siblings))
-    sibling_sp_edgelist = torch.tensor([(from_node, to_node, dist)
+    sibling_sp_edgelist = torch.tensor([(int(from_node), int(to_node), dist)
                                         for from_node, dct in sps
                                         for to_node, dist in dct.items()],
                                        dtype=torch.long)