d-ailin / GDN

Implementation code for the paper "Graph Neural Network-Based Anomaly Detection in Multivariate Time Series" (AAAI 2021)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Results of other version of libraries

windows7897 opened this issue · comments

Hi,
Thank you for the open source code.

I reproduce your model with the same libraries with other version

torch 1.13.0
torch-cluster 1.6.0+pt113cu116
torch-geometric 2.2.0
torch-scatter 2.1.0+pt113cu116
torch-sparse 0.6.16+pt113cu116
torch-spline-conv 1.2.1+pt113cu116
torchmetrics 0.11.4
torchvision 0.14.0

from my case, i get flowing results
F1 Recall Precision
SWAT 0.778047 0.643728 0.982768
WADI 0.461867 0.349515 0.682105

There is huge deviation at WADI dataset

I can't doubt any cause except library version

I want to make up the environment with your same case
pytorch-geometric can't be installed on pytorch version 1.5.1 now

so would you check your model performance at recent version of libraries?

Sorry for the late reply. It might be a bit troublesome to adapt the codes with recent version of libraries as there seems changes in some API or implementations in the updated libraries. As for the performance, could you check if the data is processed like https://github.com/d-ailin/GDN/tree/main/scripts? It could also probably affect the result. Thanks.

Thank you for your reply

My data have been preprocessed by your code

Could you share your environment by Docker images?