Link to the GIFT-EVAL benchmark: Github Repo Paper
- Follow installation instructions in the GIFT-EVAL repository to download the dataset from its huggingface dataset repository.
- Also, set the environment variable
GIFT_EVAL
to the path where the dataset is downloaded.
echo "GIFT_EVAL=/path/to/gift-eval" >> .env
- Quick start example:
python run.py --config config/default/mean.yaml \
--seed_everything 0 \
--model.forecaster.init_args.mode batch \
--data.data_manager.init_args.dataset gift/ett1/H/long \
--data.data_manager.init_args.path ./datasets \
--trainer.default_root_dir ./exps
Note
The dataset name for the GIFT-EVAL format should be specified as follows: "gift/" + "dataset_name (main_name/freq)" + "short/medium/long"
. For example, gift/ett1/H/long
. More dataset names can be found in the GIFT-EVAL repository (for example naive.ipynb).