Questions about paper and implementation consistency #3

New issue

Open

opened 2025-10-14 17:12:34 -06:00 by navan · 0 comments

navan commented

2025-10-14 17:12:34 -06:00

Owner

Originally created by @TeddyGlass on 11/11/2020

I read your paper and it was very interesting.
I have some questions about your paper and implementation consistency.

The first question is about the numerical formula (3) in your paper.
In your implementation, dose the formula (3) corresponds to update function defined as the aggregation of vectors along the edges in the file of train.py (line 48)?

def update(self, matrix, vectors, layer):
        hidden_vectors = torch.relu(self.W_fingerprint[layer](vectors))
        return hidden_vectors + torch.matmul(matrix, hidden_vectors)

If then, sigmoid function is not used in line of the return process, though the formula (3) describes sigmoid.

The second question is about formula (5, 6) in your paper.
The update of the edge information through the activation function is described in this formula (5, 6), but this process dose not seems to be implemented. Edge information is only used for extraction fingerprint based on r-radius subgraph.

Is the implementation of molecularGNN in this repository a simpler version of your paper?

Thank you

*Originally created by @TeddyGlass on 11/11/2020* I read your paper and it was very interesting. I have some questions about your paper and implementation consistency. The first question is about the numerical formula (3) in your paper. In your implementation, dose the formula (3) corresponds to update function defined as the aggregation of vectors along the edges in the file of train.py (line 48)? ```python def update(self, matrix, vectors, layer): hidden_vectors = torch.relu(self.W_fingerprint[layer](vectors)) return hidden_vectors + torch.matmul(matrix, hidden_vectors) ``` If then, sigmoid function is not used in line of the return process, though the formula (3) describes sigmoid. The second question is about formula (5, 6) in your paper. The update of the edge information through the activation function is described in this formula (5, 6), but this process dose not seems to be implemented. Edge information is only used for extraction fingerprint based on r-radius subgraph. Is the implementation of molecularGNN in this repository a simpler version of your paper? Thank you

No labels

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference: github/molecularGNN_smiles#3

No description provided.

Rows
Columns