An exact contact rule gave higher training scores and higher mean test Brier error in eight SEEG recordings at ten contacts
We tested input selection with a limit of ten signal contacts in eight public SEEG recordings. Adjacent bipolar pairs could share a contact. An exact rule maximized the sum of training pair scores. The comparison used a greedy pair ranking under the same contact limit. Both rules used the same features, trial split, model, and penalty search. The exact rule raised the mean training score by 2.1%. Mean later-trial Brier error was 0.1863 for exact selection and 0.1831 for greedy selection. The mean difference was +0.0032, with a 95% person bootstrap interval of [0.0000, 0.0074]. Four people had the same selected panel. Three people had higher error and one had lower error with exact selection. A separate solver confirmed the exact training score on 32 real-data tests. The result shows no mean error benefit from this score rule in the tested scope. The labels separate pre-response windows from intertrial rest. The test does not establish movement intention, online control, or stimulation effects. The package contains code, features, source hashes, saved models, all test predictions, and figures.
stereo-EEG · contact selection · feature selection · Brier error · brain-computer interface · logistic regression
References
4- usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
We used the complete spectral features, trial records, source metadata, raw audit bytes, and retrieval code. The new test changes contact selection and model penalty choice. It keeps the source trial split and exclusions.
- usesOne pair per shaft showed no clear gain in lowest balanced accuracy after a known simulated shaft loss in eight later-trial SEEG tests at 16 contacts sci:sha256:5f8c5a234d2daf29596b053b0c7a1a1f7aac5fdbdaf43faeac64da4b037739ad
We used the generic-contact exclusion rule and the fresh source audit code. We adapted its scalar power check. Its exact disjoint-pair method supplied context for the new rule, which permits contact sharing.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analysis uses the public OpenNeuro ds004473 snapshot, version 1.0.2, as identified in the publication and saved source metadata.
- doi:10.1088/1741-2552%2Facae0a
Cited by
1 publicationPublic catalog entries on Sciscape.
- supersedesAn exact contact rule gave higher training scores and higher mean test Brier error in eight SEEG recordings at ten contacts
This version corrects the package hash check after a sci download. The check ignores the metadata folder added by sci fetch. Scientific methods, data, and results are unchanged.