17e60a48de
NaN initial features propagate through the PPO policy network and produce garbage position sizing on the very first inference call. Replace with 0.0 so the first prediction is safe (neutral). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
adaptive-strategy
Adaptive trading strategy framework combining ensemble ML models, execution algorithms, market microstructure analysis, and risk management.
Overview
AdaptiveStrategy is the top-level type. It wraps an EnsembleCoordinator in Arc<RwLock> and coordinates predictions from four model types -- DQN, PPO, TFT, and Mamba2 -- loaded via InferenceAdapterBridge from the ml crate.
EnsembleConfig is re-exported from ml (not defined in this crate).
Modules
- config -- Strategy configuration, seeded IDs (
default-production,development,aggressive) - ensemble --
EnsembleCoordinator, model weighting, signal aggregation - execution -- TWAP, VWAP, Implementation Shortfall, POV, Arrival Price algorithms
- microstructure -- Order book analysis, trade flow classification, price impact modeling
- regime -- Market regime detection (HMM, threshold, ML-based)
- risk -- Position sizing (Kelly, risk parity, vol targeting), portfolio limits, drawdown monitoring
Features
| Cargo feature | Description |
|---|---|
postgres |
Hot-reload strategy config from PostgreSQL |
Usage
use adaptive_strategy::{AdaptiveStrategy, StrategyConfig};
let config = StrategyConfig::default();
let strategy = AdaptiveStrategy::new(config).await?;
Testing
SQLX_OFFLINE=true cargo test -p adaptive-strategy --lib
Dependencies
Core: tokio, candle-core, serde, tracing, chrono
ML models provided by the ml crate.