hjnormey/sentiment-engine
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

malkhut(wire): fill quality as PRIMARY optimization target

Browse Source 
Fill quality is MALKHUT's core aim. Wired end-to-end:

1. FillQuality state (state.py):
   - slippage_bps, price_improvement_bps, levels_consumed
   - is_maker_fill, rolling_fill_rate, post_fill_adverse_bps
   - fill_value_score: composite metric for optimization
   - Added to MarketWorldState.fill_quality field

2. HftBacktestCWM.transition() (hft_cwm.py):
   - _compute_fill_quality() computes all metrics per transition
   - Fill quality now tracked for every CWM step
   - Empty book guards added for safety

3. MinimalCryptoLOBCWM.transition() (core.py):
   - Same fill quality computation for deterministic fallback
   - Empty book guards added

4. Reward function (hft_cwm.py):
   - fill_quality_reward = w_fill_probability * fill_value_score (PRIMARY)
   - Bonus for maker fills that improve price
   - Penalty for adverse selection after fill
   - Base reward (PnL, adverse selection, fees) preserved

5. PerformanceMatrix (selector.py):
   - RegimeStrategyScore: 4 new fill quality fields
   - record(): accepts fill_rate, slippage, price_improvement, fill_value_score
   - EMA updates for all fill quality metrics

6. EpisodeResult (cma_trainer.py):
   - avg_fill_value_score, avg_price_improvement_bps, avg_post_fill_adverse_bps
   - Accumulated per-step during _run_episode
   - Recorded to PerformanceMatrix in evaluate_candidate

All 1379+ tests green.
This commit is contained in:
Codex
2026-07-15 15:22:25 +02:00
parent fa76070c79
commit 618ad723e3
5 changed files with 284 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
MALKHUT/malkhut/training/selector.py
View File

@@ -147,8 +147,9 @@ class RegimeClassifier:
@dataclass
class RegimeStrategyScore:
"""Performance score for a strategy in a specific regime.
"""Performance score for a strategy in a specific regime + venue.
Fill quality metrics are the PRIMARY optimization target.
Manifold fields (for Mode 2 recommendation):
confidence: 0.0-1.0, how reliable is this score
support_count: how many evaluations produced this score
@@ -165,6 +166,11 @@ class RegimeStrategyScore:
confidence: float = 1.0
support_count: int = 1
distance_to_nearest: float = 0.0
# Fill quality metrics (PRIMARY)
avg_fill_rate: float = 0.0
avg_slippage_bps: float = 0.0
avg_price_improvement_bps: float = 0.0
avg_fill_value_score: float = 0.0
class PerformanceMatrix:
@@ -192,6 +198,10 @@ class PerformanceMatrix:
drawdown_bps: float = 0.0,
adverse_fill_ratio: float = 0.0,
venue: str = "bingx",
fill_rate: float = 0.0,
slippage_bps: float = 0.0,
price_improvement_bps: float = 0.0,
fill_value_score: float = 0.0,
) -> None:
"""Record a strategy's performance in a regime on a specific venue."""
key = (regime, strategy_id, venue)
@@ -204,12 +214,20 @@ class PerformanceMatrix:
new_pnl = alpha * pnl_bps + (1 - alpha) * existing.avg_pnl_bps
new_dd = alpha * drawdown_bps + (1 - alpha) * existing.avg_drawdown_bps
new_adverse = alpha * adverse_fill_ratio + (1 - alpha) * existing.avg_adverse_fill_ratio
new_fill_rate = alpha * fill_rate + (1 - alpha) * existing.avg_fill_rate
new_slip = alpha * slippage_bps + (1 - alpha) * existing.avg_slippage_bps
new_improve = alpha * price_improvement_bps + (1 - alpha) * existing.avg_price_improvement_bps
new_fv = alpha * fill_value_score + (1 - alpha) * existing.avg_fill_value_score
else:
new_score = score
new_episodes = 1
new_pnl = pnl_bps
new_dd = drawdown_bps
new_adverse = adverse_fill_ratio
new_fill_rate = fill_rate
new_slip = slippage_bps
new_improve = price_improvement_bps
new_fv = fill_value_score
self._scores[key] = RegimeStrategyScore(
strategy_id=strategy_id,
@@ -223,6 +241,10 @@ class PerformanceMatrix:
confidence=min(1.0, new_episodes / 10.0),
support_count=new_episodes,
distance_to_nearest=0.0,
avg_fill_rate=new_fill_rate,
avg_slippage_bps=new_slip,
avg_price_improvement_bps=new_improve,
avg_fill_value_score=new_fv,
)
self._strategy_regime_history[strategy_id].append(regime)