How to add extra edges to graph data?

[songsong0425]

Hi, I'm trying to build a homogeneous graph for GCN.
During the process, I'd like to expand the Data by adding extra edges.
For the initial graph construction, I ran the below code:

# Generate Data object by allocating edge_index and edge_type
edge_index = torch.zeros((2, len(mapped_relation_df)), dtype=torch.long)
edge_type = torch.zeros(len(mapped_relation_df), dtype=torch.long)

for i in range(len(mapped_relation_df)):
    edge_index[0,i] = mapped_relation_df.iloc[i][0]
    edge_index[1,i] = mapped_relation_df.iloc[i][2]
    edge_type[i] = mapped_relation_df.iloc[i][1]
    
homo_g = Data(edge_index=edge_index, edge_type=edge_type)
homo_g.num_nodes = len(entity_dict)
homo_g
#Data(edge_index=[2, 2159252], edge_type=[2159252], num_nodes=47824)

After that, I want to combine other graph data by connecting the edge index.

df
#         A        B        A_1        B_1
# 0       11736    322      47824      47884
# 1       15205    14489    47825      47867
# ...	...	...	...	...	...
# 1078    8808     5342     47907      47939
# 1079    19208    1881     47921      47956

A and B columns are matched id in homo_g and A_1 and B_1 are new indices for the future node features.
In other words, edge_index[0] contains A and B, and edge_index[1] contains A_1 and B_1.
As a result, I expect the data like below figure:

[wsad1]

Could you clarify this please? Do you want to add edges represented by A_1 and B_1 to your Data object?

[songsong0425]

Sorry for unclear description. Yes. As you said, I want to add edges A_1 and B_1 to Data (i.e., homo_g) object.
During hours, I tried the below code:

df_mappedid = torch.tensor(np.hstack(df[['A', 'B']].values.tolist()))
df_newid = torch.tensor(np.hstack(df[['A_1', 'B_1']].values.tolist()))
df_combinedid = torch.stack((df_mappedid, df_newid ))

edge_index_combine = torch.cat([homo_g.edge_index, df_combinedid], dim=1)    
final_graph = Data(edge_index=edge_index_combine, x=features)

But I guess that there are other simpler approaches.

[wsad1]

I think the approach you shared is the right one.

[songsong0425]

Sorry for rate reply and thank you for your comfirmation!