Displayed reward ratios showed no clear decrease in later choice loss with a learned probability state in a fixed ds005364 split
We tested whether displayed reward ratios improve choice prediction with a learned probability state. We used 95 task records from OpenNeuro ds005364 version 1.0.0. The fixed split had 48 training people and 47 test people. We fitted models on 5,809 choices and tested them on 3,025 later choices. The reward gain was -0.001297 natural-log units per choice. The 95% bootstrap interval was [-0.011903, 0.007397]. The result did not show a clear reduction in loss. Each of 2,000 bootstrap draws repeated learning-rate selection and model fitting. The five-trial model from prior work gave the same result as before. This study extends known results on the same data. The package preserves source files, hashes, code, predictions, sampled people, and repeat-run checks.
reinforcement learning · choice prediction · reward ratios · delta rule · log loss · behavioral modeling · 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
Uses its source files, parser, person split, convex fitting functions, and published five-trial comparison.
- usesdoi:10.18112/openneuro.ds005364.v1.0.0
Source task event files and dataset metadata used in the fixed split.
- doi:10.1038/s41598-024-69772-y
- doi:10.7554/eLife.49547
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesThe choice forecast gain interval for full versus five trial probability filters included zero in a fixed ds005364 split
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.