runica_c

C implementation of runica for ICA decomposition of EEG data.

Build

Osx

make

Windows

gcc -c src/runica_simple.c -o obj/runica_simple.o -O3 -Isrc
gcc -c src/rtGetInf.c -o obj/rtGetInf.o -O3 -Isrc
gcc -c src/rtGetNaN.c -o obj/rtGetNaN.o -O3 -Isrc
gcc -c src/rt_nonfinite.c -o obj/rt_nonfinite.o -O3 -Isrc
gcc -c main.c -o obj/main.o -O3 -Isrc
gcc -o runica.exe obj/runica_simple.o obj/rtGetInf.o obj/rtGetNaN.o obj/rt_nonfinite.o obj/main.o -lm

Run

# Run ICA
./run_runica.sh [dataset] [nchans] [suffix]
./run_runica.sh ./data/eeglab_data 32 _darwin

# Plot topographies (requires MATLAB/EEGLAB)
./run_plotica.sh [dataset] [nchans] [suffix]

# Compare platforms
./run_compica.sh [dataset] [suffix1] [suffix2]
./run_compica.sh ./data/eeglab_data _darwin _win

Output files with platform suffix:

• dataset.wts_suffix - unmixing matrix (32×32 float64)
• dataset.sph_suffix - sphering matrix (32×32 float64)

Current Implementation

MATLAB Coder generated code with hardcoded parameters:

• Algorithm: Standard Infomax ICA (logistic)
• Extended ICA: ❌ Not implemented (would require tanh for sub-Gaussian)
• PCA: ❌ Not implemented
• maxsteps: 512 (hardcoded)
• stop: ~1e-6 (hardcoded)
• Channels: 32 (hardcoded)
• Random seed: Fixed (deterministic results)

To enable configurable parameters (extended, PCA, stop, maxsteps), see test/HOWTO_PARAMETERIZE.md.

Testing

cd test
./run_all_tests.sh _darwin    # Run all tests
./test_basic.sh _darwin        # Basic ICA test
./test_reproducibility.sh      # Verify deterministic results
./test_crossplatform.sh _darwin _win  # Compare platforms

See test/README.md for details.

Requirements

• macOS with Accelerate framework (or Linux with BLAS)
• gcc compiler
• MATLAB with EEGLAB (for plotting and comparison)