DeepAlpha v11.0 — 70.9% Walk-Forward Accuracy

DeepAlpha v11.0

Highlights

• 70.9% walk-forward validated accuracy on out-of-sample data
• 72 ML features engineered for crypto market microstructure
• LightGBM + XGBoost ensemble with automatic model selection
• Walk-forward validation with anchored expanding window

New in v11.0

• Temporal Fusion Transformer (TFT) module for attention-based forecasting
• TransformerGRU hybrid architecture for sequence modeling
• HMM regime detection (Hidden Markov Model) for market state classification
• Graph Neural Network module for cross-asset correlation learning
• 7 new feature engineering modules
• Improved signal calibration with dynamic thresholds

Architecture

• Multi-model ensemble (LightGBM, XGBoost, TFT, TransformerGRU)
• 72 features: OHLCV, orderbook imbalance, funding rates, volatility surfaces, sentiment
• Walk-forward validation with purged cross-validation
• Risk management: dynamic position sizing, drawdown limits, correlation filters

Supported Exchange

• Bybit (perpetual futures via v5 API)

Installation

pip install deepalpha

Links

• Website: https://deepalphabot.com
• Documentation: https://deepalphabot.com/docs
• PyPI: https://pypi.org/project/deepalpha/