TSMA

usage

1. Environmental preparation

To install the required Python dependencies, run:

pip install -r requirements.txt

For reproducibility, we also packaged the conda environment. You can [download] it and run tar -xzf TSMA.tar.gz -C ~/.conda/envs/.

The experiments are conducted using Python 3.9 and ​CUDA 12.8​. Other versions may be compatible but are not officially tested.

2. Download Datasets

Please download and extract the data into ./dataset.

• Supervised training Datasets from TSLib : [Download].
• Large-scale pre-training
  • ERA5-Family form OPenLTM : [Download].
  • UTSD form OPenLTM: [Download].

3. training

   run

python run.py --config config/seq96_patch48_ETTh1_TSMA.yaml

Acknowledgment

We’re deeply grateful to the following GitHub repositories for their outstanding code and the effort their maintainers have invested.

• OpenLTM(https://github.com/thuml/OpenLTM)
• TSLib(https://github.com/thuml/Time-Series-Library)