Deep BCI study

To get the data, please send an email to [email protected]

Installation

• Install conda : https://conda.io/docs/user-guide/install/index.html
• Install PyTorch for your platform (CUDA not necessary) : https://pytorch.org/

$ git clone <this repo>
$ cd deep_bci_study
$ conda env create
$ source activate deep_bci
$ pip install -r requirements.txt

Run

Data pre-rocessing

Pre-process the data with the following commands :

#RSVP dataset, 48 sensors (dont forget to edit load_step_1.py before)
python load_step_1.py --cache-dir /tmp/cache --output-dir /<DIR>/rsvp48
python load_step_2.py --input-dir /processed_data/rsvp48

#MEEG dataset, 48 sensors (dont forget to edit load_step_1.py before)
python load_step_1.py --cache-dir /tmp/cache --output-dir /<DIR>/meeg48
python load_step_2.py --input-dir /processed_data/rsvp48

#RSVP dataset, 48 sensors, TOPO format (dont forget to edit load_step_1.py before)
python load_step_1.py --cache-dir /tmp/cache --output-dir /<DIR>/rsvp48_topo
python load_step_2.py --input-dir /processed_data/rsvp48_topo
...

Generate the scenario files

python gen_scenario.py --base-path /processed_data --output-dir /scenarios

Run the models

Run one scenario (one subject) with one model and show test results (it's convenient for testing a model):

python deep_run_and_test.py --scenario /scenarios/rsvp48/VPfat.json --model 'mlp'

Run one scenario (one subject) with one model (no testing) :

python deep_run.py --scenario /scenarios/rsvp48/VPfat.json --model 'mlp'

Run one scenario for all subjects (run sequentially). Start by editing launch.py then :

#check with
python launch.py
#execute with
python launch.py --execute

Run the models on the local cluster

Connect to the local cluster:

ssh 10.69.111.81

Clone this gitlab repo and install all deps on the cluster. Data must be put in /mnt/data/<your_name>/ directory.

Run one scenario for all subjects on the local cluster. Start by editing launch.py then :

#check with
python launch.py --no-cluster
#execute with
python launch.py --no-cluster | sh

Scenarios

Models description