Official implementation of the paper.

Install PyTorch first, then:

pip install -r requirements.txt

All datasets are expected in the dataset/ directory. The 10 datasets used in the paper are:

These can be sourced from Time-Series-Library.

Full CAPS model:

bash caps.sh

LinAttn + RoPE ablation baseline:

bash caps_baseline.sh

Metrics are logged to Weights & Biases. Set WANDB_MODE=disabled to run offline.