Releases: proccl/quantforecast
Releases · proccl/quantforecast
v1.1.1
v1.1.0 - Walk-forward CV & Modular Architecture
What's New in v1.1.0
🔬 Optimization
- Switched from Bayesian optimization to Walk-forward Cross-Validation via Optuna
- Best params: d_model=64, n_heads=8, n_layers=3, seq_len=20
- Test accuracy: 60.0%
🏗️ Architecture Refactor
- Reorganized into modular
src/&scripts/structure - Added config management, unified data loader, and backtest engine
📊 Backtest Enhancements
- Backtest chart now shows a large header with: timestamp + T+1 signal + target price
- Live data updates via akshare + Sina/Tencent real-time APIs
📈 Latest Results (2026-04-14)
| Metric | Value |
|---|---|
| Strategy Return | +3.11% |
| Buy & Hold | -11.70% |
| Excess Return | +14.81% |
| Max Drawdown | -5.63% |
| Sharpe | 0.92 |
| Test Accuracy | 60.0% |
⚠️ Known Issues
- Regression output volatility is still high; numerical stability needs further tuning