Full covariance had lower shape loss than a model with one fitted shared factor in 36 ds004100 SEEG patient means at 1–80 Hz and weight 0.5 after test scale fitting

We compared full covariance with a model that has one fitted shared factor. Both models kept the same shaft blocks. The data were two interictal SEEG records for each of 36 patients in OpenNeuro ds004100 version 1.1.3. The primary test used 1 to 80 Hz signals and mixture weight 0.5. Each model fit one test power scale. Full covariance had lower shape loss in all 36 patient means and 71 of 72 record directions. The mean gain was 0.029504 nats per contrast dimension. The 95 percent patient bootstrap interval was 0.023822 to 0.035518. The fitted factor also improved on shaft blocks alone. All six fixed sensitivity settings had a positive mean full-over-factor gain. The package preserves input matrices, fitted loadings, code, patient scores, source hashes, and reproduction instructions. Fresh public bytes rebuilt all 72 covariance records exactly. The full repeat from rebuilt data matched the saved results. This is a new analysis of a known cohort. The fit checks do not prove a global optimum. The result tests one shared statistical factor. It does not establish causal neural connections or clinical benefit.

SEEG · intracranial EEG · covariance · shared factor · epilepsy · covariance modeling · shape loss · cross-validation