gitadmin/foxhunt
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

feat(reward-v7): add cea_weight, order_credit_weight, risk_efficiency_weight config fields

Browse Source 
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
jgrusewski
2026-04-07 20:11:34 +02:00
parent 457dff3db7
commit b615559507
1 changed files with 26 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
crates/ml/src/trainers/dqn/config.rs
View File

@@ -772,6 +772,19 @@ pub struct DQNHyperparameters {
pub dd_threshold: f64,
/// Asymmetric loss scaling factor (prospect theory, default 1.5)
pub loss_aversion: f64,
/// Reward v7: Counterfactual Exposure Advantage weight.
/// Blends per-branch advantage signal into the scalar reward.
/// 0.0 = disabled, 1.0 = full advantage signal.
pub cea_weight: f64,
/// Reward v7: Order type microstructure credit weight.
/// Rewards cost-efficient execution (LimitMaker > Market).
pub order_credit_weight: f64,
/// Reward v7: Intra-trade risk efficiency weight.
/// Rewards clean winning trades with minimal intra-trade drawdown.
pub risk_efficiency_weight: f64,
/// Position staleness rent per step held
pub time_decay_rate: f64,
/// Q-value gap threshold for trade conviction filter.
@@ -1485,12 +1498,14 @@ impl DQNHyperparameters {
self.feature_mask_fraction = (self.feature_mask_fraction * di).clamp(0.0, 0.8);
self.feature_noise_scale *= di;
// Loss shaping family
// Loss shaping family — v7: scale CEA + order credit + risk efficiency
let li = self.loss_shaping_intensity;
self.asymmetric_dd_weight *= li;
self.regret_blend = (self.regret_blend * li).clamp(0.0, 1.0);
self.trade_clustering_penalty *= li;
self.cea_weight = (self.cea_weight * li).clamp(0.0, 2.0);
self.order_credit_weight = (self.order_credit_weight * li).clamp(0.0, 1.0);
self.risk_efficiency_weight = (self.risk_efficiency_weight * li).clamp(0.0, 1.0);
self.position_entropy_weight *= li;
// v6 legacy (zeroed defaults, no-op)
self.asymmetric_dd_weight *= li;
// Ensemble family
let ei = self.ensemble_intensity;
@@ -1532,7 +1547,7 @@ impl DQNHyperparameters {
w_dd: 1.0,
w_idle: 0.01,
dd_threshold: 0.02,
loss_aversion: 1.5,
loss_aversion: 1.0,
time_decay_rate: 0.0005,
q_gap_threshold: 0.05, // Tier 2 default: mild conviction gating (hyperopt searches [0.0, 0.5])
huber_delta: 100.0, // BUG #12 FIX: Scale delta 100x for gradient explosion fix (was 1.0)
@@ -1589,9 +1604,13 @@ impl DQNHyperparameters {
feature_mask_fraction: 0.3, // #23: mask 30% of features each epoch
feature_noise_scale: 0.1, // #22: add N(0, 0.1*std) noise to features
time_reversal_mod: 5, // #11: 20% of episodes played backwards
regret_blend: 0.3, // #17: 30% regret + 70% raw PnL
regret_blend: 0.0, // #17: zeroed in v7 (no-op, kept for TOML compat)
position_entropy_weight: 0.01, // #19: reward += 0.01 * H(position_histogram)
trade_clustering_penalty: 0.05, // #25: penalize temporally clustered trades
trade_clustering_penalty: 0.0, // #25: zeroed in v7 (no-op, kept for TOML compat)
// Reward v7 — Counterfactual Branch Attribution
cea_weight: 0.3,
order_credit_weight: 0.1,
risk_efficiency_weight: 0.1,
pruning_epoch: 50, // #20: compute mask at epoch 50
pruning_fraction: 0.7, // #20: prune 70% of smallest weights
bottleneck_dim: 2, // #31: 2D information compression (gem of gems)