Adds curiosity-driven exploration (ForwardDynamicsModel + ICM reward), configurable branching DQN fields (num_order_types, num_urgency_levels), regime-conditional importance sampling with ADX/CUSUM thresholds, and GPU curiosity training kernel support. Also fixes remaining 5 clippy errors from WIP merge: - gpu_smoketest: add 8 missing DQNConfig fields - curiosity.rs: replace needless_range_loop with slice fill - benchmark files: remove redundant #[cfg_attr] on unconditionally ignored tests 40 files changed, +1486/-1068 lines. 0 clippy errors, 0 warnings. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
15 KiB
15 KiB