Code for "A Multi-label Classification Approach to Increase Expressivity of EMG-based Gesture Recognition" by Niklas Smedemark-Margulies, Yunus Bicer, Elifnur Sunger, Stephanie Naufel, Tales Imbiriba, Eugene Tunik, Deniz Erdoğmuş, and Mathew Yarossi

Setup and Usage

Use make to create python environment, install dependencies, and install pre-commit hooks.

To reproduce our experiments:

1. Download the dataset from Zenodo using bash scripts/fetch_dataset.sh
2. Preprocess data to extract features using python scripts/preprocess_data.py

• To check that all data have been downloaded and preprocessed successfully, use pytest to run test_load_data_dict

3. Run experiments. Note:

• DRY_RUN=True at the beginning of each script causes jobs to be printed, but not executed; change it to DRY_RUN=False to actually launch.
• Each script is a large grid of jobs, and all results are simple stored flat in one folder.
• To run fewer jobs at a time, truncate the list of jobs in the script.

python scripts/run_experiment_1.py
python scripts/run_experiment_2.py
python scripts/run_experiment_3.py

4. Create plots. Note:

• The list of expected results is found from each experiment launch script. If some results are missing (e.g. due to runs that timed-out or failed), they will be counted and printed to a text file.
• Failed runs can be re-launched using python scripts/rerun_failed path/to/missing.txt, by providing the path to this text file of missing runs.

python scripts/process_experiment_1.py
python scripts/process_experiment_2.py
python scripts/process_experiment_3.py
python multi_label_emg/feature_similarity.py --doubles_method subset_uniform --frac_doubles_per_class 0.005 --singles_method none --gamma median

PDF

To read our paper, see: https://arxiv.org/pdf/2309.12217.pdf

Dataset

To use our dataset, see: https://zenodo.org/records/10393194

Citation

If you use this code or dataset, please use one of the citations below.

Article citation:

@article{smedemark2023multi,
    title={A Multi-label Classification Approach to Increase Expressivity of EMG-based Gesture Recognition},
    author={
        Smedemark-Margulies, Niklas and 
        Bicer, Yunus and
        Sunger, Elifnur and
        Naufel, Stephanie and
        Imbiriba, Tales and
        Tunik, Eugene and
        Erdo{\u{g}}mu{\c{s}}, Deniz and
        Yarossi, Mathew
    },
    journal={arXiv preprint arXiv:2309.12217},
    year={2023},
    month={09},
    day={13},
    url={https://arxiv.org/abs/2309.12217},
}

Dataset citation:

@dataset{smedemarkmargulies_2023_10393194,
  title={{EMG from Combination Gestures with Ground-truth Joystick Labels}},
  author={
    Smedemark-Margulies, Niklas and 
    Bicer, Yunus and
    Sunger, Elifnur and
    Naufel, Stephanie and
    Imbiriba, Tales and
    Tunik, Eugene and
    Erdo{\u{g}}mu{\c{s}}, Deniz and
    Yarossi, Mathew
  },
  year={2023},
  month={12},
  day={15},
  publisher={Zenodo},
  version={1.0.1},
  doi={10.5281/zenodo.10393194},
  url={https://zenodo.org/records/10393194},
}