fix(rl): conf gate floor 0.001 → 0.0 so controller can fully disable
When C51 LCB is ≈0 for all actions (typical early training), even 0.001 threshold blocks everything. Floor=0.0 lets the controller drive to zero when trades dry up, then ratchet back as Q calibrates. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
@@ -1617,7 +1617,7 @@ impl IntegratedTrainer {
|
||||
(crate::rl::isv_slots::RL_OUTCOME_ALPHA_INDEX, 0.1),
|
||||
(crate::rl::isv_slots::RL_GATE_WARMUP_STEPS_INDEX, 10000.0),
|
||||
(crate::rl::isv_slots::RL_GATE_DONES_TARGET_INDEX, 0.02),
|
||||
(crate::rl::isv_slots::RL_GATE_CONF_MIN_INDEX, 0.001),
|
||||
(crate::rl::isv_slots::RL_GATE_CONF_MIN_INDEX, 0.0),
|
||||
(crate::rl::isv_slots::RL_GATE_CONF_MAX_INDEX, 0.50),
|
||||
(crate::rl::isv_slots::RL_GATE_FRD_MIN_INDEX, 0.05),
|
||||
(crate::rl::isv_slots::RL_GATE_FRD_MAX_INDEX, 0.50),
|
||||
|
||||
Reference in New Issue
Block a user