gitadmin/foxhunt
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
2,140 Commits 13 Branches 0 Tags
2245e347c9034112e45c4241237d820cfcb53fae
Commit Graph

5 Commits

Author SHA1 Message Date
jgrusewski
58f4f26113 refactor(ml): extract regime-detection, explainability, paper-trading 
- ml-regime-detection (1.2K lines): feature_classifier, hmm modules.
  Depends on ml-core + ml-dqn (RegimeType). 23 tests passing.

- ml-explainability (329 lines): integrated_gradients module.
  Depends on ml-core + candle-core. 4 tests passing.

- ml-paper-trading (389 lines): broker, pnl_tracker modules.
  Depends on ml-ensemble (TradeAction, TradeSignal). 8 tests passing.

Total: 21 sub-crates extracted from ml monolith.
ml reduced from ~260K to ~90K lines (65% extracted).
All tests: 841 ml + 35 in new sub-crates = 876 passing.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
 2026-03-08 15:17:22 +01:00
jgrusewski
4e0225e090 feat(ml): BF16 VarBuilder, checkpoints, and training tensors for PPO 
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
 2026-03-03 16:16:07 +01:00
jgrusewski
a41387c3f9 fix(ml): make IG completeness axiom test deterministic 
Use a purely linear network (no ReLU) for the completeness axiom test.
IG on linear functions is mathematically exact, so the test is
deterministic regardless of random weight initialization. Tighten
tolerance from 20% to 1% (f32 rounding only). Keep ReLU network in
the basic test for non-linear verification.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
 2026-03-02 15:41:23 +01:00
jgrusewski
2e15d1d834 fix: trading_service PPO type mismatch and IG test tolerance 
- trading_service: PPO predict() used TradingAction match but act() now
  returns FactoredAction. Use target_exposure() mapped to 0-1 range.
- IG completeness axiom test: relax tolerance from 5% to 20% (random
  weights with ReLU non-linearity and trapezoidal rule can exceed 5%).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
 2026-03-02 15:41:23 +01:00
jgrusewski
764deb7ddb feat(ml): add integrated gradients for feature importance 
Implements IntegratedGradients using Candle autograd (Var::from_tensor +
backward). Computes attributions by integrating input gradients along
interpolation path from baseline to input.

Verified via completeness axiom test: sum(attributions) ≈ F(x) - F(baseline).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
 2026-03-02 15:41:23 +01:00