82b595f633
Replace #error guards in common_device_functions.cuh with safe defaults (STATE_DIM=72, MARKET_DIM=42, PORTFOLIO_DIM=8). Add MAX_STATE_DIM=128 for safe stack array sizing. Add runtime parameters to kernels that used compile-time #define: - attention_kernel.cu: state_dim, num_heads - attention_backward_kernel.cu: state_dim, num_heads - backtest_forward_ppo_kernel.cu: state_dim, num_actions - backtest_metrics_kernel.cu: num_actions, order_actions, urgency_actions - dqn_utility_kernels.cu: state_dim - her_relabel_kernel.cu: state_dim - iql_value_kernel.cu: state_dim (all 3 kernel functions) - iqn_dual_head_kernel.cu: state_dim - monitoring_kernel.cu: num_actions, order_actions, urgency_actions This enables single-cubin-per-kernel precompilation — no #define injection needed at runtime. 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;