search

Ljiajie / Addgraph

The code of Addgraph frame.
20 stars 7 forks source link

Issue for the method of anomaly generation #3

Open mgao97 opened 1 year ago

mgao97 commented 1 year ago

Hey!

I am confused about the way how to generate anomaly edges for each snapshot and how to get the dimension output of the H_list of node embedding like this "torch.Size([10, 12503, 50])".

Any answers related to the above two questions are helpful to me. Thanks a lot!

Ljiajie commented 1 year ago

Hey!

I am confused about the way how to generate anomaly edges for each snapshot and how to get the dimension output of the H_list of node embedding like this "torch.Size([10, 12503, 50])".

Any answers related to the above two questions are helpful to me. Thanks a lot!

Hi! The generation of anomaly edges is based on the method of negative sampling in paper: Knowledge graph embedding by translating on hyperplanes. H_list is used to collect the output hidden state of each snapshots, so the initial shape is [w(window size of attention), nodes, hidden_dim]. Over time, the first dimension will gradually increase from 1 to w + (number of snapshots).

mgao97 commented 1 year ago

Rover that, thank you for your quick answer.

---Original--- From: @.> Date: Tue, Dec 6, 2022 22:06 PM To: @.>; Cc: @.**@.>; Subject: Re: [Ljiajie/Addgraph] Issue for the method of anomaly generation(Issue #3)

Hi! The generation of anomaly edges is based on the method of negative sampling in paper: Knowledge graph embedding by translating on hyperplanes. H_list is used to collect the output hidden state of each snapshots, so the initial shape is [w(window size of attention), nodes, hidden_dim]. Over time, the first dimension will gradually increase from 1 to w + (number of snapshots).

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