tindzk / SelfOrganisingMaps

Stuart's self-organising map models

SelfOrganisingMaps

Implementation of GCAL model reported in Stevens et al., (2013) J. Neurosci. paper.

First install morphologica and Abseil, then build in the usual cmake way:

mkdir build
cd build
cmake -DCMAKE_INSTALL_PREFIX=/usr/local ..
make
cd ..

/usr/local refers to the path where morphologica and other custom dependencies were installed.

Then run model using e.g.:

export LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH
build/sim/gcal configs/config.json --seed=1 --mode=1 --input=2

The final 3 numbers are:

1. random seed
2. Mode -- 0: displays off, 1: displays on
3. Training pattern type -- 0: oriented Gaussians, 1: preloaded vectors*, 2: video camera input

*Note that if using preloaded vectors you will need to supply a hdf5 file as the "patterns" parameter in the JSON file. An example can be generated as follows:

cd configs/
python -m venv venv
source venv/bin/activate
pip install -r  requirements.txt
python genTestPatterns.py

This creates the file configs/testPatterns.h5.

If the path to file of a saved weightfile is optionally appended, these weights will be used, else initial weights are random.

Enjoy!

Troubleshooting

hdf5 not found

If cmake fails because it could not find hdf5, the pkg-config file hdf5.pc may be missing:

-- Have pkg-config, searching for libmorphologica...
-- Checking for module 'libmorphologica'
--   Package 'hdf5', required by 'libmorphologica', not found

This can be solved by copying the version-tagged .pc file to hdf5.pc. Example:

cp /usr/lib/pkgconfig/hdf5-1.10.5.pc $CMAKE_INSTALL_PREFIX/lib/pkgconfig/hdf5.pc