A simple democratic way to decode classes using bayes rule.
The underlying algorithm is based on ideas from a neuroscience paper,
Insanally, M. N. et al. Spike-timing-dependent ensemble encoding by non-classically responsive cortical neurons. eLife 8, (2019).
We put these same ideas to use, with a more general focus. The big idea is still very simple.
To train the decoder we,
- build a kernel probability dist,
- tune its bandwidth by CV,
- use the result to do a sequential bayesian decode, where each channels dist gets a 'vote'.
To test, this scheme gets repeated, minus tuning, for rolling time windows.
That's it.
TODO
git clone pip install .
- scikit learn
- standard conda install