Ensemble Training - Crypto Sentiment

⚙️ Ensemble Training Configuration

Select Cryptocurrencies to Train

BTC (Bitcoin) ETH (Ethereum) SOL (Solana)

Prediction Horizons (hours)

4 hours (Tactical) 8 hours (Trend Strength) 24 hours (Strategic)

ℹ️ Simplified to 3 horizons only (4h, 8h, 24h) for better signal quality

🔬 Optimization Settings

Optuna Trials Number of Bayesian optimization trials (more = better but slower)

Validation Days Days used for weight optimization validation

📅 Training Data Range

Start Date (optional) Leave empty to use "Training Days" below

End Date (optional) Leave empty to use latest available data

💡 Tip: Use date range to train on specific periods and avoid bias in backtesting. For example, train on Jan-Mar data, then backtest on April.

Training Data Days (fallback) Used only if Start/End dates are not specified. (365 = 1 year, 730 = 2 years)

Optimization Metric Trading metric used for Optuna optimization

📊 Model Selection

Skip Technical Models

Skip training of technical models (RF, XGB, GB, Ridge). Check to train only sentiment models.

Skip Sentiment Models

Skip training of sentiment models. Check to train only technical models.

Force Model Replacement

Replace all old models after training, even if new models are not better. Useful when changing the entire logic. Models are replaced AFTER training completes.

💡 Tip: Uncheck both to train full ensemble (technical + sentiment)

📊 Model Backtest

Test ensemble model performance on historical data with realistic trading simulation.

Select Cryptocurrencies to Backtest

BTC (Bitcoin) ETH (Ethereum) SOL (Solana)

Period (days)

Capital ($)

⚙️ Trading Parameters

Stop-Loss (%)

Take-Profit (%)

Max Position (%)

Min Confidence

Min Conviction

📊 What Gets Trained

3 Horizons: 4h (tactical), 8h (trend), 24h (strategic)
Multi-Model: 4 models per horizon (RF, XGB, GB, Ridge)
Technical Features: 60+ indicators (RSI, MACD, Bollinger, etc.)
Regression Mode: Predicts exact % price change
Optuna: Hyperparameter + ensemble weight optimization
Trading Metrics: Sharpe, expectancy, profit factor

⏱️ Estimated Time

Quick Test: 15-20 minutes (1 symbol, 1 horizon, 20 trials)
Medium: 1-2 hours (1 symbol, 3 horizons, 50 trials)
Full Production: 4-6 hours (3 symbols, 3 horizons, 50 trials)
High Quality: 12-18 hours (3 symbols, 3 horizons, 200 trials)
Maximum Quality: 30-40 hours (3 symbols, 3 horizons, 500 trials)
Per Model: ~15-20 min (50 trials), ~60-80 min (200 trials), ~3h (500 trials)

🎯 Expected Improvements

Multi-Horizon: 3 horizons (4h, 8h, 24h) - reduced noise
Trading Bot: Weighted ensemble (50% 24h, 35% 8h, 15% 4h)
Strategic Alignment: 24h + 8h must agree for trade
Position Sizing: 0-35% capital based on conviction

🎯 Ensemble ML System Training

📊 Current Training Status