search

d-ailin / GDN

Implementation code for the paper "Graph Neural Network-Based Anomaly Detection in Multivariate Time Series" (AAAI 2021)
MIT License
488 stars 142 forks source link

multi-step prediction with iteration #90

Closed FortunateStarstar86 closed 8 months ago

FortunateStarstar86 commented 8 months ago

I'm sorry to bother you with a problem I can't solve. when I manipulate the input iteratively by appending the predicted value at T time step, and use the new input to predict the values at T+1,the Losses have increased exponentially. 32333f5eb5cd604889847e35d54e53c I used the trained model test data (test.csv) to output "predicted", which resulted in exponentially larger predictions from the fourth round. Can you give some analysis or advice? f3bce4f16e0f5a4bfbed8e5bc10d02c Looking forward to your reply!

FortunateStarstar86 commented 8 months ago

the second pic: f3bce4f16e0f5a4bfbed8e5bc10d02c

d-ailin commented 8 months ago

Hi, i see. I guess it might be due to the error propagation in the sequential modeling, but it seems a bit severe from your cases. In this case, you might want to use more time steps as input, or try the way of training with an output of 10 dimensions. Sorry for the inconvenience.

FortunateStarstar86 commented 8 months ago

Hi, i see. I guess it might be due to the error propagation in the sequential modeling, but it seems a bit severe from your cases. In this case, you might want to use more time steps as input, or try the way of training with an output of 10 dimensions. Sorry for the inconvenience.

OK，Thanks for your reply！