Online SSVEP in Riemanian manifold

Pre-requisite

Follow Timeflux Getting Started Instructions.

Installation

You have the choice between virtual env or conda env:

• conda env

     conda env create -f  environment.yaml
     conda activate ssvep-env
  

• virtual env

     python3 -m venv venv
     source venv/bin/activate
     pip install -r requirements.txt
  

Usage

• Script to get HDF5 (todo: parse args to allow user to choose params)

    python make_hdf5.py
  

  • this should download data from MOABB and convert them in HDF5 timeflux-replayable data that will be stored in folder ./data
  • By default, train_runs=('run_0',), ie. the first run will be considered as calibration data and the rest as test data. You can change those if you wish.

• Replay the data and online SSVEP fit_predict with Timeflux

    timeflux -d graphs/main.yaml
  

  • This will replay data from one subject 'in real time' (by default, subject 12) /!\ takes between 9 and 18 minutes to run - as long as the experiment /!\
  • If you want to try an choose the subject to replay, set variable FILE in your environment, or launch the command with -e FILE={subject-number-between-1-and-12}.
    For example, to replay data from subject #1, run :

    timeflux -e FILE=1 -d graphs/main.yaml
  

  • This should display events and predictions in the console.

  • The output events will be dumped in a csv with name set line 14 of file graphs/dump.yaml : predictions_{FILE}.csv.

  • Output looks like :

  label	data	timestamp
  train_starts	{}	2020-01-01 00:01:08.703125
  flickering_starts	{'target': '13Hz'}	2020-01-01 00:01:08.707031250
  flickering_starts	{'target': '13Hz'}	2020-01-01 00:01:17.707031250
  flickering_starts	{'target': '17Hz'}	2020-01-01 00:01:53.707031250
  flickering_starts	{'target': '21Hz'}	2020-01-01 00:02:02.707031250
  ...	...	...
  train_stops	{}	2020-01-01 00:05:53.621093750
  flickering_starts	{'target': '13Hz'}	2020-01-01 00:06:39.941406250
  predict	{'result': '13Hz'}	2020-01-01 00:06:38.941406250
  flickering_starts	{'target': '13Hz'}	2020-01-01 00:06:48.941406250
  predict	{'result': '13Hz'}	2020-01-01 00:06:47.941406250

  

• Notebooks

    jupyter notebook
  

  • "Online-Offline.ipynb" illustrate the sklearn pipeline offline
  • "Timeflux prediction.ipynb" loads the csv output file from Timeflux with events and predictions and plot a confusion matrix

References

• data: MOABB/SSVEPExo dataset from E. Kalunga PhD in University of Versailles [1]_. (url). (classes = rest, 13Hz, 17Hz, 21Hz)
• matlab implementation: https://github.com/emmanuelkalunga/Online-SSVEP
• paper SSVEP: https://hal.archives-ouvertes.fr/hal-01351623/document
• paper RPF: ttps://hal.archives-ouvertes.fr/hal-02015909/document

Results