Evaluation on recorded EEG#
Six independently runnable lessons use explicit small EEGDashDataset subsets.
The SSVEP lessons download approximately 21.1 MB for three participants;
the session and MOABB lessons use about 11 MB of real motor imagery.
Set EEGDASH_CACHE_DIR to reuse the signals. All scores come from held-out
recorded trials and their observed event labels.
plot_50_within_subject_evaluation.py: hold out complete trials within each known participant.plot_51_cross_subject_evaluation.py: leave one participant out.plot_52_cross_session_evaluation.py: transfer between genuine sessions from one participant.plot_53_learning_curves.py: add nested training-subject subsets while keeping one validation participant fixed.plot_54_compare_two_pipelines.py: compare paired subject scores and explain the limited resolution of a three-participant statistical test.plot_55_moabb_interop.py: pass EEGDash-loaded signals into MOABB’s dataset interface and obtain real labelled epochs (requiresmoabb).
Choose the split unit to match deployment, fit learned transformations only on training data, and preserve trial identities when creating windows. These small subsets demonstrate evaluation mechanics, not benchmark claims.
