f39219be9e
Root cause: catastrophic OOS eval (Sharpe -280) traced to 11 issues including hardcoded training dynamics and per-bar CPU inference. Search space (dqn.rs): - Add warmup_ratio [0.0, 0.15] — was hardcoded to 0 - Add lr_decay_type [Constant/Linear/Cosine] — was hardcoded Constant - Add min_epochs_before_stopping [2, 6] — was 1000 (disabled) - Add minimum_profit_factor [1.1, 2.0] — was hardcoded 1.5 - Widen entropy_coefficient [0.01, 0.2] for 45-action factored space GPU-batched eval (evaluate_baseline.rs): - 1024-bar chunked inference for both DQN and PPO - DQN: batch_greedy_actions per chunk (was per-bar select_action) - PPO: action_probabilities + GPU argmax per chunk - ~1000x fewer GPU kernel launches - Add trade_sharpe_ratio for hyperopt-comparable metric Preprocessing (preprocessing.rs): - Add compute_clip_bounds/clip_outliers_with_bounds for leakage-free clipping across train/val/test splits Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
ml
10-model ML ensemble for the Foxhunt HFT system, built on Candle v0.9.1.
Models
- DQN (Rainbow) — deep Q-network with prioritized replay, dueling heads, noisy nets
- PPO — proximal policy optimization with GAE, LSTM policies, clip-higher
- TFT — temporal fusion transformer for multi-horizon forecasting
- Mamba2 — state space model for sequence prediction
- Liquid Networks — biologically inspired networks for non-stationary data
- TLOB — transformer-based limit order book analysis
- KAN — Kolmogorov-Arnold networks
- xLSTM — extended LSTM architecture
- TGGN — temporal graph neural network
- Diffusion — diffusion-based generative model
Key Modules
ensemble— model ensemble coordination and confidence aggregationhyperopt— PSO-based hyperparameter optimization with per-model adapterstrainers— unified training loops (DQN, PPO, supervised)inference—InferenceAdaptertrait for predictioncheckpoint— model checkpointing and restorationevaluation— walk-forward evaluation pipeline
Usage
use ml::dqn::DQN;
use ml::ppo::PpoTrainer;