Can't use SVM for decoding

unfoldtoolbox / UnfoldDecode.jl

MIT License

1 stars 4 forks source link

Can't use SVM for decoding #4

Open geigermartin opened 6 months ago

geigermartin commented 6 months ago

Hi!

While going through the tutorial overlapcorrectedLDA_options.jl, I tried to exchange the LDA for an SVM classifier. However, this unfortunately is not working (even though the SVM is MLJ supported as well).

My code:

using MLJLIBSVMInterface
SVC = @load SVC pkg = LIBSVM
uf_svm = fit(UnfoldDecodingModel,des,evt,dat,SVC(),"eventA"=>:condition;
    nfolds=2,
    UnfoldFitkwargs=(;solver=customsolver),
    eventcolumn=:event,
    multithreading = false)
plot_erp(coeftable(uf_svm))

behinger commented 2 months ago

ok this is working partially now. The problem is coeftable which requires the LDA output format, nt sure why this is incompatible with SVM output frmat. Don't have time to fiix it right now - someone with more experience in MLJ could fix this super fast i think