From f05a4c5d447752db2a00ae50cfac6e05538e31e4 Mon Sep 17 00:00:00 2001 From: jgrusewski Date: Tue, 7 Apr 2026 21:02:47 +0200 Subject: [PATCH] config: enable position entropy (0.02) + add hyperopt search range Anti-Flat-collapse: rewards exposure diversity via per-episode position visit histogram entropy bonus. Co-Authored-By: Claude Opus 4.6 (1M context) --- config/training/dqn-hyperopt.toml | 1 + config/training/dqn-localdev.toml | 2 +- config/training/dqn-production.toml | 2 +- config/training/dqn-smoketest.toml | 2 +- 4 files changed, 4 insertions(+), 3 deletions(-) diff --git a/config/training/dqn-hyperopt.toml b/config/training/dqn-hyperopt.toml index 7e05f64e6..c83ede1f5 100644 --- a/config/training/dqn-hyperopt.toml +++ b/config/training/dqn-hyperopt.toml @@ -78,6 +78,7 @@ w_idle = [0.0, 0.1] dd_threshold = [0.005, 0.03] # HFT: tight drawdown tolerance (0.5%-3%) loss_aversion = [1.0, 1.0] time_decay_rate = [0.0001, 0.005] +position_entropy_weight = [0.01, 0.1] # Trade conviction filter q_gap_threshold = [0.0, 0.5] # 0.0=trade every bar, 0.5=high conviction only diff --git a/config/training/dqn-localdev.toml b/config/training/dqn-localdev.toml index 6a0baf743..7b9e0764b 100644 --- a/config/training/dqn-localdev.toml +++ b/config/training/dqn-localdev.toml @@ -94,7 +94,7 @@ stochastic_depth_prob = 0.2 asymmetric_dd_weight = 0.5 time_reversal_mod = 5 regret_blend = 0.3 -position_entropy_weight = 0.01 +position_entropy_weight = 0.02 trade_clustering_penalty = 0.05 ensemble_disagreement_penalty = 0.3 mixup_alpha = 0.2 diff --git a/config/training/dqn-production.toml b/config/training/dqn-production.toml index 7887969ed..29d65673d 100644 --- a/config/training/dqn-production.toml +++ b/config/training/dqn-production.toml @@ -105,7 +105,7 @@ stochastic_depth_prob = 0.2 asymmetric_dd_weight = 0.5 time_reversal_mod = 5 regret_blend = 0.3 -position_entropy_weight = 0.01 +position_entropy_weight = 0.02 trade_clustering_penalty = 0.05 ensemble_disagreement_penalty = 0.3 mixup_alpha = 0.2 diff --git a/config/training/dqn-smoketest.toml b/config/training/dqn-smoketest.toml index 5801c1b0d..fd2406d70 100644 --- a/config/training/dqn-smoketest.toml +++ b/config/training/dqn-smoketest.toml @@ -96,7 +96,7 @@ trade_clustering_penalty = 0.02 # 0.02 vs production 0.05 ensemble_disagreement_penalty = 0.1 # 0.1 vs production 0.3 regret_blend = 0.15 # 0.15 vs production 0.3 time_reversal_mod = 10 # 10% reversed vs production 20% -position_entropy_weight = 0.005 # Half of production +position_entropy_weight = 0.02 causal_weight = 0.05 # Half of production causal_intervention_interval = 50 # Less frequent than production (10) bottleneck_dim = 0 # Disabled for smoketest (tiny network can't afford bottleneck)