967c964dd8
The CUDA kernels (.cu files) were updated to accept 4 branch pointers and b3_size, but the Rust launch functions still only passed 3 branches. This caused CUDA_ERROR_INVALID_VALUE on H100 due to argument count mismatch. Fixed 7 launch sites in gpu_dqn_trainer.rs: - launch_c51_loss: added on_b3/tg_b3/on_next_b3 pointers + b3_i32 - launch_c51_mixup: added branch_3_size arg - launch_c51_grad: added b3_i32, fixed thread count 3*na → 4*na - launch_mse_loss: added on_b3/tg_b3/on_next_b3 pointers + b3_i32 - launch_mse_grad_inner: added b3_i32, fixed thread count 3*na → 4*na - apply_cql_gradient: added b3_i32 - compute_expected_q: added b3 Also updated max_branch calculations to include b3. 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;