Jiffy is a lightweight CNN that produces a data-dependent embedding for a time series dataset. Follow the testing instructions below to reproduce the paper's results.
We test the implementation on 6 publicly available datasets:
- arabic_digits: access here
- AUSLAN: access here
- character_trajectories: access here
- libras: access here
- ecg: access here
- wafer: access here
But there's no need to download them all separately! Create the datasets directory by running the following script in the root directory:
./download_datasets.sh
You can then run the evaluation of Jiffy on all 6 datasets with the following command:
python models/test_scripts/test_jiffy.py