OpenBCI_Python library: best practices for integrating custom BCI algorithms

phora · April 2015

I've been trying to write a program to manually label which waveforms are sad or happy since the OpenBCI GUI doesn't provide a simple way to add manual markers to the data. The source for it can be found here: manually_labeled_waveforms.py

What I should be expecting is a bunch of packets with varying labels next to them. Instead, I get an empty file and a bunch of messages along the line of "skipped packet with ID#".

I haven't been able to reproduce the same problem when I ran user.py in the OpenBCI_Python repo when I run the user.py file with these steps:

python2 user.py -p /dev/ttyUSB0 -a csv_collect
Then in the program, type /start
Wait for some data to come on. You will never see the message, "I skipped packet with ID#" over and over again no matter what you do.
Then type /stop
Then type /exit
You should get a csv file filled with EEG data

What'm I doing wrong with invoking the library to connect to the board in my program? I decided to try writing a python program to handle classifying the EEG waveforms as a fallback since the project is in a time crunch.

phora · April 2015

Forgot to link back to the OpenBCI_Python repository since I typed it out the previous post in a hurry: https://github.com/OpenBCI/OpenBCI_Python/blob/master/user.py

wjcroft · April 2015

Phora, hi.

I'm not directly answering your Python library question. But there is another way to record timing mark data along with an EEG stream, using the LSL 'apps' LabRecorder and Keyboard. The resulting combined streams are then recorded into an XDF file by LabRecorder.

http://code.google.com/p/labstreaminglayer/wiki/LabRecorder

http://code.google.com/p/labstreaminglayer/
https://github.com/dmedine/labstreaminglayer/ (migration to GitHub by UCSD)

http://code.google.com/p/labstreaminglayer/wiki/TimeSynchronization

http://openbci.com/forum/index.php?p=/discussion/334/labstreaminglayer-into-bcilab-eeglab-matlab

I have not used this myself, but it is being used in many other BCI lab settings.

----

Jeremy @jfrey or Rodrigo @Rceballos98 , may have more comments on your library question.

William

phora · April 2015

@wjcroft: While I highly appreciate the answer (and it would be awesome if I just needed to label them), I'd need to use the code above as a basis to classify the EEG waveforms on a Rubix A-10 board (the fallback plan in case coding the classifications on the Arduino or the RFDuino failed) to ID the emotions we're targeting before sending them to the Arduino board below controlling the cart. The reason why I'm looking at the Python library for OpenBCI is:

It's easier for me to work with since I'm familiar with python and the usual libraries for working with signals (numpy, scipy)
It's cross-platform, meaning I can develop the classifying software and test it on my desktop before putting it on the board

wjcroft · April 2015

Makes sense, did not realize you were doing live classification. Here's another thread with some aBCI links,

http://openbci.com/forum/index.php?p=/discussion/325/functions-linked-to-10-20-locations-abci-affective-bci

Probably easiest to run your app on a board separate from OpenBCI, as you say embedded Linux platforms are ubiquitous at this stage.

http://linuxgizmos.com/ringing-in-2015-with-40-linux-friendly-hacker-sbcs/

wjcroft · April 2015

@phora , on further thought, why not just follow the modular nature of what Jeremy @jfrey and Rodrigo have put together with Python, and make your code / classifier a "plugin" or element of the existing framework?

In other words, instead of trying to chop out sections of the OpenBCI Python tree... Just use the whole tree and incorporate your mods as additions to the data flow already there. The user.py may be setting up the environment for the rest of the library to work.

http://openbci.com/forum/index.php?p=/discussion/233/github-com-openbci-python

William

jfrey · April 2015

Hello,

As William said, it should be easier to maintain your program if you package it as a plugin; for example the "user.py" script handles the daisy module and is programmed with the right init sequence ("svd") -- both of which seem missing in your code. Also in the near future I should backport the "emergency reset" from OpenViBE implementation and make recordings more reliable. I believe the "print" plugin is simple enough so you can start from here, but don't hesitate to ask for help if you have trouble to implement your own modifications -- I tried to make things scalable but the real stress test begins once someone else is extending your code

Eitan · May 2015

Hi All

I have tried to use python API for OPENBCI, with little success.

I took a look at the OpenBCI_Python at github but still I a have no success.

1) How do i get a streaming of the board and read the data?

2) I found a command to control the filter, but is it the 50hz or the 60hz and how I activate the other?

3) I will be happy to see an example for P300 application using python API.

Best

Eitan

wjcroft · May 2015

@Eitan I've merged your questions into this existing thread on Python. Check out some of the earlier posts, and also the thread below with instructions on how to start the stream. You did not find the instructions in the README adequate to start the stream? Can you be more specific?

http://openbci.com/forum/index.php?p=/discussion/233/github-com-openbci-python

If you use the Search box in the right column for 'P300' you can see some other threads; Jeremy @jfrey has done a OpenViBE tutorial, etc.

This post elaborates on the "/start" and related commands,

http://openbci.com/forum/index.php?p=/discussion/comment/1913/#Comment_1913

OpenBCI_Python library: best practices for integrating custom BCI algorithms

Comments