Skip to content

Latest commit

 

History

History
121 lines (95 loc) · 4.09 KB

readme.md

File metadata and controls

121 lines (95 loc) · 4.09 KB

r.bci

Overview

This is a small project to build an R-based free+open-source cross-platform GUI tool for brain-computer interface research, like that of EEGLAB or BCILAB, to accomplish the main tasks of data science therein (adapted from Ricky Ho's article on the subject):

  1. Data acquisition
  2. Data visualization
  3. Preprocessing algorithm design
  4. Classification
  5. Report generation

Other goals

  • Specific, well-defined formats and data structures for EEG data (especially taking advantage of data.table)
  • Cross-platform support through GTK
  • Improved visualizations (adding to the already excellent ones in R)
  • Increased collaboration and reproducibility through free+open software

Features

  • Modules for each of the above tasks, and more:
    • Import
      • MATLAB
      • RData
      • CSV
    • Filtering
    • Downsampling
    • Exploration
      • Histograms
      • Summary stats
      • Grand means
      • Data partitioning
    • Transforms
      • PCA
      • CSP
    • Unsupervised learning
      • k-means
    • Supervised learning
      • SDA
      • SVM
      • Naive Bayes
    • Report generation
      • Automatic code generation
      • Literate programming (with knitr)

Current status

v0.1: Usable alpha reached. (See development milestones for more.)

How to build and run

  • Note: that this is experimental software and has only been tested on 64-bit Linux. If you have problems, please include a traceback() and your system information in an issue.

Requirements:

  • You should have the latest version of R (tested on Pumpkin Helmet, amd64 Linux).
  • Make sure you have a copy of GTK2, including development headers (libgtk2.0-dev for Linux users)
  • For MPI-based parallelism, you will need OpenMPI development headers and binaries (libopenmpi-dev, openmpi-bin)
  • May also require curl as RCurl depends on it (specifically, libcurl4-gnutls-dev).

Running rBCI

The rest of the packages should install through R when run:

git clone https://github.com/talexand/rbci.git
cd rbci
R
source('./gwidgets/init_interface.R')

Why not just use *LAB?

  1. EEGLAB and BCILAB contain a lot of great work that represents the combined effort of many highly skilled researchers, and they are extremely easy to use in a properly configured software/hardware environment. However, both toolboxes depend heavily on a proprietary platform, namely MATLAB, whose idiosyncratic syntax, large footprint, and high price (2000 USD as of this writing) present hurdles to learning, building, and applying ideas and techniques, both new and old, to a fast-growing research area. An R implementation has the potential to help connect a large portion of the data science/machine learning community to the field of BCI and even accelerate things by taking advantage of the vast number of actively developed, freely available software packages (I'm looking at you, expensive MATLAB Parallel Toolbox.)

  2. It's true that there are already a number of great GUI libraries for data science in R, for example, Rattle. However, after two years of research, I have found none aimed squarely at the task types and data structures that are ubiquitous in, for example, EEG research. (A great example of this is epoched multivariate time series.)

  3. EEGLAB and BCILAB don't support parallelism or distributed computing.

Roadmap

See the issues page for tickets marked 'enhancement'.

Wishlist

  • stream input for online algorithms from BCI2000 (incorporating rstreamBCI)
  • Complete replication of most popular algorithms from EEGLAB, BCILAB, SIFT, etc.
  • Core code written entirely in fast precompiled languages (C, C++)
  • Interface to wavelet tools
  • Additional report generation features like
    • AUC analysis in feature space
  • Modular tool format so that interface components can be swapped in and out of chain (like bci2000)