4249ebc961
Sentinel = 0.0 per pearl_first_observation_bootstrap. First observation of each EMA replaces the sentinel directly, no blending. Slots [510..520): - loss_cap (510): adaptive loss cap for reward clamp - alpha_ema (511): Wiener-α EMA for loss_cap producer - wr_ema (512): win-rate EMA driving loss_cap adaptive ramp - hold_cost_scale (513): hold penalty cost multiplier - target_hold_pct (514): hold-engagement target - hold_pct_ema (515): hold-engagement EMA - hold_reward_ema (516): hold-action reward EMA - n_step (517): multi-step TD horizon adapter - aux_conf_threshold (518): auxiliary task confidence threshold - aux_gate_temp (519): auxiliary task gating temperature
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;