FRandR count maps gave lower mean later-date log loss than duration maps in seven sleep SEEG patients

We tested whether past event duration improved a forecast of later FRandR counts across SEEG pairs. FRandR is the supplied label for a fast ripple that occurs with a ripple. We used public OpenNeuro ds003498 version 1.0.1. The seven selected patients had depth electrodes in the source paper and at least two recording dates. We made count and duration maps from the earlier half of each patient's dates. We tested 18,415 marks on 15 later patient dates. Both maps used the same pair set and one percent uniform smoothing. The equal patient mean log loss was 1.982170 for counts and 2.110595 for duration weights. The mean gain was count loss minus duration loss: -0.128424 natural log units per target mark. Three of seven patient gains were positive. The 95 percent patient t interval was [-0.409830, +0.152982]. It requires independent, approximately normal patient gains. The exact two-sided sign-flip value was 0.1875 under independent patients and symmetric null gains. One patient accounted for most of the mean difference. All eight fixed checks had negative mean gains, with t intervals that included zero. Every valid FRandR mark shared its supplied boundaries with a ripple mark. The BIDS files call the depth channels ECOG; we preserve this metadata conflict. This reused cohort cannot establish a population effect, neural duration mechanism, or clinical accuracy. The package preserves 655 source files, hashes, code, event features, maps, results, figures, and repeat instructions. Independent source and score checks passed. A fresh source download reproduced all 18 main output files byte for byte.

high frequency oscillations · FRandR · stereoelectroencephalography · event duration · forecasting · log loss · sleep · epilepsy