A Python package to create real-time Brain Computer Interfaces (BCI's). Data synchronisation and pipelining handled by the Lab Streaming Layer, machine learning with Pytorch, scikit-learn or TensorFlow, leveraging packages like AntroPy, SciPy and NumPy for generic time and/or frequency based feature extraction or optionally have the users own custom feature extraction class used.
The goal of PyBCI is to enable quick iteration when creating pipelines for testing human machine and brain computer interfaces, namely testing applied data processing and feature extraction techniques on custom machine learning models. Training the BCI requires LSL enabled devices and an LSL marker stream for timing stimuli.
All the examples found on the github not in a dedicated folder have a pseudo LSL data generator enabled by default, createPseudoDevice=True
so the examples can run without the need of LSL capable hardware. Any generic LSLViewer can be used to view the generated data, example viewers found on this link.
If samples have been collected previously and model made the user can set the clf
, model
, or torchModel
to their sklearn, tensorflow or pytorch classifier and immediately set bci.TestMode()
.
Examples of supported LSL hardware here!
- Add optional LSL outlet configuration for class estimator (either send every classification or send on classification change - help with reducing spam on classification if estimator time is very short)
- Add example showing previously saved models.
- Add example showing how feature data can be saved and used to build models so model creation can be done offline whilst data collection and classification can be done online.
- Update and verify via appveyor for 3.12.
For stable releases use: pip install pybci-package
For development versions use: pip install git+https://github.com/LMBooth/pybci.git
or
git clone https://github.com/LMBooth/pybci.git
cd pybci
pip install -e .
Or optionally, install and run in a virtual environment:
Windows:
python -m venv my_env
.\my_env\Scripts\Activate
pip install pybci-package # For stable releases
# OR
pip install git+https://github.com/LMBooth/pybci.git # For development version
Linux/MaxOS:
python3 -m venv my_env
source my_env/bin/activate
pip install pybci-package # For stable releases
# OR
pip install git+https://github.com/LMBooth/pybci.git # For development version
If you are not using windows then there is a prerequisite stipulated on the pylsl repository to obtain a liblsl shared library. See the liblsl repo documentation for more information.
Once the liblsl library has been downloaded pip install pybci-package
should work.
(currently using pybci-package due to pybci having name too similar with another package on pypi, issue here.)
There has been issues raised with Linux successfully running all pytests and examples, there is a dockerfile included in the root repository outlining what should be a successful build of ubuntu 22:04.
There is an Ubuntu 22.04 setup found in the Dockerfile in the root of the directory which can be used in conjunction with docker.
Once docker is installed call the following in the root directory:
sudo docker build -t pybci .
sudo docker run -it -p 4000:8080 pybci
Then either run the pybci
CLI command or run pytest Tests
to verify functionality.
Download the Dockerfile and run
After installing pybci and downloading and extracting the pybci git repository, navigate to the extracted location and run pip install requirements-devel.txt
to install pytest, then call pytest -vv -s Tests\
to run all the automated tests and ensure all 10 tests pass (should take approximately 15 mins to complete), this will ensure pybci functionality is as desired.
Tested on Python 3.9, 3.10 & 3.11 (appveyor.yml)
The following package versions define the minimum supported by PyBCI, also defined in setup.py:
"pylsl>=1.16.1",
"scipy>=1.11.1",
"numpy>=1.24.3",
"antropy>=0.1.6",
"tensorflow>=2.13.0",
"scikit-learn>=1.3.0",
"torch>=2.0.1"
Earlier packages may work but are not guaranteed to be supported.
import time
from pybci import PyBCI
if __name__ == '__main__':
bci = PyBCI(createPseudoDevice=True) # set default epoch timing, looks for first available lsl marker stream and all data streams
while not bci.connected: # check to see if lsl marker and datastream are available
bci.Connect()
time.sleep(1)
bci.TrainMode() # now both marker and datastreams available start training on received epochs
accuracy = 0
try:
while(True):
currentMarkers = bci.ReceivedMarkerCount() # check to see how many received epochs, if markers sent to close together will be ignored till done processing
time.sleep(0.5) # wait for marker updates
print("Markers received: " + str(currentMarkers) +" Accuracy: " + str(round(accuracy,2)), end=" \r")
if len(currentMarkers) > 1: # check there is more then one marker type received
if min([currentMarkers[key][1] for key in currentMarkers]) > bci.minimumEpochsRequired:
classInfo = bci.CurrentClassifierInfo() # hangs if called too early
accuracy = classInfo["accuracy"]
if min([currentMarkers[key][1] for key in currentMarkers]) > bci.minimumEpochsRequired+10:
bci.TestMode()
break
while True:
markerGuess = bci.CurrentClassifierMarkerGuess() # when in test mode only y_pred returned
guess = [key for key, value in currentMarkers.items() if value[0] == markerGuess]
print("Current marker estimation: " + str(guess), end=" \r")
time.sleep(0.2)
except KeyboardInterrupt: # allow user to break while loop
print("\nLoop interrupted by user.")
PyBCI is a python brain computer interface software designed to receive a varying number, be it singular or multiple, Lab Streaming Layer enabled data streams. An understanding of time-series data analysis, the lab streaming layer protocol, and machine learning techniques are a must to integrate innovative ideas with this interface.
An LSL marker stream is required to train the model, where a received marker epochs the data received on the accepted datastreams based on a configurable time window around set markers - where custom marker strings can optionally have their epoch time-window split and overlapped to count as more then one marker, example: in training mode a baseline marker may have one marker sent for a 60 second window, whereas target actions may only be ~0.5s long, when testing the model and data is constantly analysed it would be desirable to standardise the window length, we do this by splitting the 60s window after the received baseline marker in to ~0.5s windows. PyBCI allows optional overlapping of time windows to try to account for potential missed signal patterns/aliasing - as a rule of thumb it would be advised when testing a model to have a time window overlap >= 50% (Shannon-Nyquist criterion). See here for more information on epoch timing.
Once the data has been epoched it is sent for feature extraction, there is a general feature extraction class which can be configured for general time and/or frequency analysis based features, ideal for data stream types like "EEG" and "EMG". Since data analysis, preprocessing and feature extraction trechniques can vary greatly between device data inputs, a custom feature extraction class can be created for each data stream maker type. See here for more information on feature extraction.
Finally a passable pytorch, sklearn or tensorflow classifier can be given to the bci class, once a defined number of epochs have been obtained for each received epoch/marker type the classifier can begin to fit the model. It's advised to use bci.ReceivedMarkerCount() to get the number of received training epochs received, once the min num epochs received of each type is >= pybci.minimumEpochsRequired (default 10 of each epoch) the model will begin to fit. Once fit the classifier info can be queried with CurrentClassifierInfo, this returns the model used and accuracy. If enough epochs are received or high enough accuracy is obtained TestMode() can be called. Once in test mode you can query what pybci estimates the current bci epoch is(typically baseline is used for no state). Review the examples for sklearn and model implementations.