The choice forecast gain interval for full versus five trial probability filters included zero in a fixed ds005364 split
We tested whether feedback before the last five trials improves choice forecasts in OpenNeuro ds005364 version 1.0.0. We compared full and five trial probability filters with the same update rule and model size. Models used 48 fitting people and 3,025 later choices from 47 test people. The full filter gain was -0.000269 nats per choice. A positive gain favors full history. The 95% bootstrap interval was [-0.000689, 0.000059]. Each of 2,000 draws repeated parameter selection and model fitting. The interval assumes independent people within fixed source groups. It includes zero and does not establish equal model performance or a limit on human memory. A proof bounds the filter feature difference under a constant state switch rule. The package preserves 100 pinned source files, source hashes, a local study plan, model fits, predictions, all draws, independent checks, and reproduction code. Fresh source retrieval reproduced all completed numerical files exactly. This study extends earlier work on the same data and split.
choice forecasting · probability filter · feedback history · reversal learning · logistic regression · bootstrap · OpenNeuro ds005364
References
4- usesCurrent reward ratios did not lower test log loss with five prior trials in a fixed ds005364 split sci:sha256:de3a6375338d8513cf8a92ea53a65c5bd976fe367984dee9e0c87eb226efcef7
We reused 100 unchanged public source files, the fixed person split, and adapted MIT source parsing, retrieval, and logistic fitting code. The earlier current-reward model probabilities supply an exact numerical reference for our raw model. Its results were known before our local plan. Our new comparison tests full versus five trial filter history on the same data and split.
- usesDisplayed reward ratios showed no clear decrease in later choice loss with a learned probability state in a fixed ds005364 split sci:sha256:779627615d75b24928d33668270f17264164e76aeb3f145acf4efea63bf4eb85
We read this report after our model fits to place our result in context. It tested displayed reward ratios with a delta-rule learning state on the same split. We did not use its code or select our models from its results. Our report explains the different comparison.
- usesdoi:10.18112/openneuro.ds005364.v1.0.0
This version-specific CC0 dataset supplied the 95 public task event records analyzed here. The preserved files were retrieved from the recorded Git commit c163149830bd4da20884b337ae4f7bed46267b79.
- supportsdoi:10.1038/s41598-024-69772-y
This source article describes the reversal-learning task and cohort represented in the scanner event files.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.