Datasetpublic

wcarvalho/Multitask_Preplay_Euclidean_Heuristics

Multitask Preplay — Euclidean heuristic sweeps Parameter-sweep data for the greedy and memory-based Euclidean-distance baselines of the Multitask Preplay paper (all H1+H2 settings, with alpha/radius/p_drop/beta/fidelity/decision parameter columns intact). Best settings used in the paper: JaxMaze alpha=10; Craftax alpha=10 with mild fidelity (r=2, p=0.1) and greedy decision. Data for the paper "Multitask Preplay" (PNAS). Analysis code:… See the full description on the dataset page: https://huggingface.co/datasets/wcarvalho/Multitask_Preplay_Euclidean_Heuristics.

sourceHugging Faceupdated 3mo agoView on Hugging Face

0likes22downloads

Dataset Card

Multitask Preplay — Euclidean heuristic sweeps

Parameter-sweep data for the greedy and memory-based Euclidean-distance baselines of the Multitask Preplay paper (all H1+H2 settings, with alpha/radius/p_drop/beta/fidelity/decision parameter columns intact). Best settings used in the paper: JaxMaze alpha=10; Craftax alpha=10 with mild fidelity (r=2, p=0.1) and greedy decision.

Data for the paper "Multitask Preplay" (PNAS). Analysis code: https://github.com/wcarvalho/multitask_preplay (branch pnas).

Splits: jaxmaze, shortcut, craftax. Each split is also available as a top-level parquet file.