chenxino / CAN

Coupled Attention Networks for Multivariate Time Series Anomaly Detection

This is the implementation of "CAN" based on PyTorch.

structure of the code:

• lib folder: some training methods and evaluation methods from mtad-gat-pytorch;
• models folder: specific implementation of "CAN" model;
• utils.py: method of loading data;
• train.py: train and run the model;
• run.sh: shell script for running models.

You can use source run.sh $gpu_id $dataset_name command to run the code, such as source run.sh 0 SWaT_10. The dataset can be obtained through mtad-gat-pytorch and iTrust.