e2bf655be2
1. expected_q_kernel layout mismatch: cuBLAS writes branch-major logits [Branch0: B×b0×NA | Branch1: ...] but kernel read sample-major [Sample0: 12×NA | Sample1: ...]. Fixed to use branch_logit_offset accumulator matching cuBLAS output layout. 2. Duplicate expected_q_kernel.cu deleted — single implementation in experience_kernels.cu. Trainer loads from experience_kernels.cubin. 3. Monitoring num_actions=7 (old 7-level exposure) → 9 (dir×mag = 3×3). Actions with dir*mag≥7 silently dropped from order/urgency counts, making it appear only 1 order type was active. Result: Action diversity 15/81 (18.5%) → 45/81 (55.6%), order=3/3. Co-Authored-By: Claude Opus 4.6 (1M context) <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;