Dear Author,
In the paper, you mention that the resulting 34 subgraph embeddings are concatenated to prevent information loss. However, I noticed in the code (specifically in Model/DRPreter.py, line 131) that you write the following:
x_cell = x[:, :-1, :].sum(dim=1)
As a result, the shape of x_cell is:
x_cell.shape: torch.Size([128, 34, 256]) ==> x_cell.shape: torch.Size([128, 256])
This operation seems to sum the embeddings instead of concatenating them, which appears to conflict with the description in the paper. Could you kindly clarify whether the embeddings are summed or concatenated in the final implementation, or if there may have been a misunderstanding on my part?
Best regards
Kevin
Dear Author,
In the paper, you mention that the resulting 34 subgraph embeddings are concatenated to prevent information loss. However, I noticed in the code (specifically in Model/DRPreter.py, line 131) that you write the following:
x_cell = x[:, :-1, :].sum(dim=1)
As a result, the shape of x_cell is:
x_cell.shape: torch.Size([128, 34, 256]) ==> x_cell.shape: torch.Size([128, 256])
This operation seems to sum the embeddings instead of concatenating them, which appears to conflict with the description in the paper. Could you kindly clarify whether the embeddings are summed or concatenated in the final implementation, or if there may have been a misunderstanding on my part?
Best regards
Kevin