search

pursuecong / WinGNN

3 stars 3 forks source link

Some question to reproduce the results of the paper #1

Open gray-sunflower opened 9 months ago

gray-sunflower commented 9 months ago

Hi, thanks for the amazing work. I have some questions regarding the codes:

  1. In the paper, the decoder is the concat of the source and destination nodes, whereas in the code is a dot product. If i change the decoder to concat the results seem higher by a large margin.

  2. The code in the line 35 means that it can not reach (n - w) therefore the last snapshot of the training set will not be used during training. This confuse me since i think the last snapshot should be used during training.

image

Thanks, I look forward to your response

pursuecong commented 9 months ago

I'm glad you're interested in our work, In response to your question, I would like to give the following explanation：

  1. We conducted the experiment uniformly based on the Settings used in Baseline experiment, and there may be better results. We did not try decoder.
  2. The value of n starts at 0, so it will contain the last snapshot. Thank you for your questions. If you have any other questions, you are welcome to contact us and we will make corresponding improvements.

At 2023-10-17 21:46:53, "gray-sunflower" @.***> wrote:

Hi, thanks for the amazing work. I have some question regarding the codes:

The decoder is the concat of the source and destination nodes, whereas in the code is a dot product. If i change the decoder to concat the results seem higher by a large margin.

The code in the line 35 means that it can not reach (n - w) therefore the last snapshot of the training set will not be used during training. This confuse me since i think the last snapshot should be used during training.

Thanks, I look forward to your response

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>