drl-trading-bot-dev2 / backtest_strategy.py
DRL Trading Bot
Feature: HTF Agent integration — live trading, API endpoints, UI tab
fc115d5
Raw
History Blame Contribute Delete
27.7 kB
#!/usr/bin/env python3
"""
Strategy Backtester
Replays the EXACT live trading pipeline on historical data to validate
model performance before deployment.
Usage:
python backtest_strategy.py # Backtest all assets
python backtest_strategy.py --asset BTCUSDT # Single asset
python backtest_strategy.py --days 180 # 6 months
"""
import sys
import os
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
import numpy as np
import pandas as pd
from pathlib import Path
from typing import Dict, List, Optional, Tuple
from datetime import datetime, timedelta
import json
import logging
import argparse
from stable_baselines3 import PPO
from stable_baselines3.common.vec_env import DummyVecEnv, VecNormalize
from src.features.ultimate_features import UltimateFeatureEngine
from src.features.correlation_engine import SimulatedDominanceEngine
from src.features.regime_detector import MarketRegimeDetector
from src.features.risk_manager import AdaptiveRiskManager
from src.backtest.data_loader import DataLoader
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s [%(name)s] %(levelname)s - %(message)s'
)
logger = logging.getLogger(__name__)
class StrategyBacktester:
"""
Replays the exact live trading pipeline on historical data.
Mirrors:
- Feature computation (UltimateFeatureEngine + SimulatedDominanceEngine)
- VecNormalize observation scaling
- PPO model inference
- Regime-adaptive SL/TP
- 60% trailing stops
- Min hold time + cooldown
"""
def __init__(
self,
model_path: str = './data/models/ultimate_agent.zip',
vec_norm_path: str = './data/models/ultimate_agent_vec_normalize.pkl',
initial_balance: float = 5000.0,
position_size: float = 0.25,
min_hold_bars: int = 4, # 4 hours at 1h timeframe
cooldown_bars: int = 2, # 2 hours cooldown after SL
):
self.initial_balance = initial_balance
self.position_size = position_size
self.min_hold_bars = min_hold_bars
self.cooldown_bars = cooldown_bars
# Load model
self.model = None
if Path(model_path).exists():
try:
self.model = PPO.load(model_path)
logger.info(f"✅ Model loaded from {model_path}")
except Exception as e:
logger.error(f"Failed to load model: {e}")
# Load VecNormalize
self.vec_normalize = None
if Path(vec_norm_path).exists():
try:
import gymnasium as gym
dummy_env = DummyVecEnv([lambda: gym.make('CartPole-v1')])
self.vec_normalize = VecNormalize.load(vec_norm_path, dummy_env)
self.vec_normalize.training = False
self.vec_normalize.norm_reward = False
logger.info(f"✅ VecNormalize loaded from {vec_norm_path}")
except Exception as e:
logger.warning(f"⚠️ VecNormalize load failed: {e}")
# Feature engines
self.feature_engine = UltimateFeatureEngine()
self.dominance_engine = SimulatedDominanceEngine()
self.regime_detector = MarketRegimeDetector()
self.risk_manager = AdaptiveRiskManager()
def _compute_observation(
self, df: pd.DataFrame, position: int,
position_price: float, balance: float
) -> Optional[np.ndarray]:
"""Compute observation vector matching live pipeline."""
if len(df) < 200:
return None
try:
# Features (same as live)
all_features = self.feature_engine.get_all_features(df)
dominance_features = self.dominance_engine.compute_simulated_dominance(df)
all_features.update(dominance_features)
features_df = pd.DataFrame(all_features)
features_df = features_df.fillna(0).replace([np.inf, -np.inf], 0)
for col in features_df.columns:
if features_df[col].dtype in [np.float64, np.float32]:
features_df[col] = features_df[col].clip(-10, 10)
features = features_df.values.astype(np.float32)
if len(features) < 1:
return None
last_features = features[-1].copy()
# Position info
current_price = df.iloc[-1]['close']
if position != 0 and position_price > 0:
if position == 1:
unrealized_pnl = (current_price - position_price) / position_price
else:
unrealized_pnl = (position_price - current_price) / position_price
else:
unrealized_pnl = 0.0
balance_ratio = (balance - self.initial_balance) / self.initial_balance
position_info = np.array([
float(position),
np.clip(unrealized_pnl, -0.5, 0.5),
np.clip(balance_ratio, -0.5, 0.5),
], dtype=np.float32)
observation = np.concatenate([last_features, position_info]).astype(np.float32)
# Apply VecNormalize
if self.vec_normalize is not None:
try:
obs_2d = observation.reshape(1, -1)
observation = self.vec_normalize.normalize_obs(obs_2d).flatten().astype(np.float32)
except:
pass
return observation
except Exception as e:
logger.warning(f"Feature computation failed: {e}")
return None
def _get_regime_adjustments(self, df: pd.DataFrame, direction: str) -> Tuple[float, float]:
"""Get regime-adaptive SL/TP multipliers."""
sl_mult, tp_mult = 1.0, 1.0
try:
regime_info = self.regime_detector.detect_regime(df)
regime_name = regime_info.regime.value
if regime_name == 'high_volatility':
sl_mult, tp_mult = 1.5, 1.5
elif regime_name in ('trending_up', 'trending_down'):
tp_mult = 1.5
elif regime_name == 'ranging':
sl_mult, tp_mult = 0.8, 0.8
except:
pass
return sl_mult, tp_mult
def run(self, symbol: str = 'BTC/USDT', days: int = 90) -> Dict:
"""
Run backtest on historical data.
Returns detailed performance report.
"""
logger.info(f"{'='*60}")
logger.info(f"BACKTESTING {symbol} — last {days} days")
logger.info(f"{'='*60}")
# Load data
loader = DataLoader()
end_date = datetime.now()
start_date = end_date - timedelta(days=days)
df = loader.load(
symbol=symbol,
timeframe='1h',
start_date=start_date.strftime('%Y-%m-%d'),
end_date=end_date.strftime('%Y-%m-%d'),
)
if df.empty or len(df) < 250:
logger.error(f"Insufficient data for {symbol}: {len(df)} candles")
return {"error": f"Insufficient data: {len(df)} candles"}
logger.info(f"Loaded {len(df)} candles ({df.index[0]} to {df.index[-1]})")
# Trading state
balance = self.initial_balance
position = 0 # 0=flat, 1=long, -1=short
position_price = 0.0
position_units = 0.0
sl_price = 0.0
tp_price = 0.0
highest_price = 0.0
lowest_price = 0.0
entry_bar = 0
last_loss_bar = -100
# Tracking
trades = []
equity_curve = []
max_equity = balance
# Lookback window for features (need 200+ bars)
start_idx = 200
for i in range(start_idx, len(df)):
# Current data slice (everything up to current bar)
current_df = df.iloc[:i+1].copy()
current_price = float(current_df.iloc[-1]['close'])
bars_in_position = i - entry_bar if position != 0 else 0
# ─── Check SL/TP ───────────────────────────────────────
if position != 0:
exit_trade = False
exit_reason = ""
exit_price = current_price
if position == 1: # LONG
highest_price = max(highest_price, current_price)
# Enhanced trailing stop (60% of gains)
if highest_price > position_price:
gain = highest_price - position_price
new_sl = position_price + gain * 0.6
if new_sl > sl_price:
sl_price = new_sl
if current_price <= sl_price and sl_price > 0:
exit_trade = True
exit_reason = "STOP_LOSS" if sl_price <= position_price else "TRAILING_STOP"
elif current_price >= tp_price and tp_price > 0:
exit_trade = True
exit_reason = "TAKE_PROFIT"
elif position == -1: # SHORT
lowest_price = min(lowest_price, current_price) if lowest_price > 0 else current_price
# Enhanced trailing stop (60% of gains)
if lowest_price < position_price and lowest_price > 0:
gain = position_price - lowest_price
new_sl = position_price - gain * 0.6
if new_sl < sl_price:
sl_price = new_sl
if current_price >= sl_price and sl_price > 0:
exit_trade = True
exit_reason = "STOP_LOSS" if sl_price >= position_price else "TRAILING_STOP"
elif current_price <= tp_price and tp_price > 0:
exit_trade = True
exit_reason = "TAKE_PROFIT"
if exit_trade:
# Calculate P&L
if position == 1:
pnl = (current_price - position_price) * position_units
else:
pnl = (position_price - current_price) * position_units
balance += position_price * position_units + pnl
trades.append({
'entry_bar': entry_bar,
'exit_bar': i,
'direction': 'LONG' if position == 1 else 'SHORT',
'entry_price': position_price,
'exit_price': current_price,
'pnl': pnl,
'reason': exit_reason,
'bars_held': bars_in_position,
'timestamp': str(current_df.index[-1]) if hasattr(current_df.index[-1], 'strftime') else str(i),
})
if "STOP_LOSS" in exit_reason:
last_loss_bar = i
position = 0
position_units = 0
position_price = 0
sl_price = 0
tp_price = 0
highest_price = 0
lowest_price = 0
continue
# ─── Get Model Action ─────────────────────────────────
if self.model is None:
equity_curve.append(balance)
continue
obs = self._compute_observation(current_df, position, position_price, balance)
if obs is None:
equity_curve.append(balance)
continue
try:
action, _ = self.model.predict(obs, deterministic=True)
action = int(action.item() if hasattr(action, 'item') else action)
except Exception as e:
action = 0
# ─── Apply Guards ─────────────────────────────────────
# Cooldown after loss
if action != 0 and position == 0 and (i - last_loss_bar) < self.cooldown_bars:
action = 0
# Min hold time
if action != 0 and position != 0 and bars_in_position < self.min_hold_bars:
action = 0
# ─── Execute Trade ────────────────────────────────────
if action == 1 and position <= 0: # BUY
# Close short first
if position == -1:
pnl = (position_price - current_price) * position_units
balance += position_price * position_units + pnl
trades.append({
'entry_bar': entry_bar,
'exit_bar': i,
'direction': 'SHORT',
'entry_price': position_price,
'exit_price': current_price,
'pnl': pnl,
'reason': 'SIGNAL_EXIT',
'bars_held': bars_in_position,
'timestamp': str(current_df.index[-1]) if hasattr(current_df.index[-1], 'strftime') else str(i),
})
position = 0
# Open long
if position == 0:
trade_value = balance * self.position_size
if trade_value > 10:
position_units = trade_value / current_price
position_price = current_price
position = 1
balance -= trade_value
entry_bar = i
highest_price = current_price
# Calculate SL/TP (regime-adaptive)
try:
sl_pct, tp_pct = self.risk_manager.get_adaptive_sl_tp(current_df, "long")
sl_mult, tp_mult = self._get_regime_adjustments(current_df, "long")
sl_pct *= sl_mult
tp_pct *= tp_mult
except:
sl_pct, tp_pct = 0.025, 0.05
sl_price = current_price * (1 - sl_pct)
tp_price = current_price * (1 + tp_pct)
elif action == 2 and position >= 0: # SELL
# Close long first
if position == 1:
pnl = (current_price - position_price) * position_units
balance += position_price * position_units + pnl
trades.append({
'entry_bar': entry_bar,
'exit_bar': i,
'direction': 'LONG',
'entry_price': position_price,
'exit_price': current_price,
'pnl': pnl,
'reason': 'SIGNAL_EXIT',
'bars_held': bars_in_position,
'timestamp': str(current_df.index[-1]) if hasattr(current_df.index[-1], 'strftime') else str(i),
})
position = 0
# Open short
if position == 0:
trade_value = balance * self.position_size
if trade_value > 10:
position_units = trade_value / current_price
position_price = current_price
position = -1
balance -= trade_value
entry_bar = i
lowest_price = current_price
# Calculate SL/TP (regime-adaptive)
try:
sl_pct, tp_pct = self.risk_manager.get_adaptive_sl_tp(current_df, "short")
sl_mult, tp_mult = self._get_regime_adjustments(current_df, "short")
sl_pct *= sl_mult
tp_pct *= tp_mult
except:
sl_pct, tp_pct = 0.025, 0.05
sl_price = current_price * (1 + sl_pct)
tp_price = current_price * (1 - tp_pct)
# Track equity
unrealized = 0
if position == 1:
unrealized = (current_price - position_price) * position_units
elif position == -1:
unrealized = (position_price - current_price) * position_units
equity = balance + (position_price * position_units if position != 0 else 0) + unrealized
equity_curve.append(equity)
max_equity = max(max_equity, equity)
# ─── Close any remaining position ─────────────────────
if position != 0:
final_price = float(df.iloc[-1]['close'])
if position == 1:
pnl = (final_price - position_price) * position_units
else:
pnl = (position_price - final_price) * position_units
balance += position_price * position_units + pnl
trades.append({
'entry_bar': entry_bar,
'exit_bar': len(df) - 1,
'direction': 'LONG' if position == 1 else 'SHORT',
'entry_price': position_price,
'exit_price': final_price,
'pnl': pnl,
'reason': 'END_OF_DATA',
'bars_held': len(df) - 1 - entry_bar,
'timestamp': str(df.index[-1]) if hasattr(df.index[-1], 'strftime') else 'end',
})
# ─── Calculate Metrics ────────────────────────────────
report = self._calculate_metrics(trades, equity_curve, symbol)
return report
def _calculate_metrics(self, trades: List[Dict], equity_curve: List[float], symbol: str) -> Dict:
"""Calculate comprehensive performance metrics."""
if not trades:
return {
'symbol': symbol,
'total_trades': 0,
'status': 'NO_TRADES',
'message': 'Model generated no trades',
}
# Basic metrics
pnls = [t['pnl'] for t in trades]
wins = [p for p in pnls if p > 0]
losses = [p for p in pnls if p <= 0]
total_return = (sum(pnls) / self.initial_balance) * 100
win_rate = len(wins) / len(pnls) * 100 if pnls else 0
avg_win = np.mean(wins) if wins else 0
avg_loss = abs(np.mean(losses)) if losses else 0
profit_factor = sum(wins) / abs(sum(losses)) if losses and sum(losses) != 0 else float('inf')
# Sharpe Ratio (annualized from hourly)
if len(equity_curve) > 1:
returns = pd.Series(equity_curve).pct_change().dropna()
if returns.std() > 0:
sharpe = (returns.mean() / returns.std()) * np.sqrt(8760) # Annualize from hourly
else:
sharpe = 0
else:
sharpe = 0
# Max Drawdown
if equity_curve:
peak = equity_curve[0]
max_dd = 0
for eq in equity_curve:
peak = max(peak, eq)
dd = (peak - eq) / peak if peak > 0 else 0
max_dd = max(max_dd, dd)
else:
max_dd = 0
# Win/Loss by reason
reason_stats = {}
for t in trades:
reason = t['reason']
if reason not in reason_stats:
reason_stats[reason] = {'count': 0, 'total_pnl': 0, 'wins': 0}
reason_stats[reason]['count'] += 1
reason_stats[reason]['total_pnl'] += t['pnl']
if t['pnl'] > 0:
reason_stats[reason]['wins'] += 1
# Direction analysis
long_trades = [t for t in trades if t['direction'] == 'LONG']
short_trades = [t for t in trades if t['direction'] == 'SHORT']
long_pnl = sum(t['pnl'] for t in long_trades)
short_pnl = sum(t['pnl'] for t in short_trades)
long_wins = sum(1 for t in long_trades if t['pnl'] > 0)
short_wins = sum(1 for t in short_trades if t['pnl'] > 0)
# Average hold time
avg_hold = np.mean([t['bars_held'] for t in trades]) if trades else 0
# Validation
passed = sharpe > 0.5 and win_rate > 50 and max_dd < 0.10
report = {
'symbol': symbol,
'status': 'PASSED ✅' if passed else 'FAILED ❌',
'total_trades': len(trades),
'total_return_pct': round(total_return, 2),
'total_pnl': round(sum(pnls), 2),
'win_rate_pct': round(win_rate, 1),
'wins': len(wins),
'losses': len(losses),
'avg_win': round(avg_win, 2),
'avg_loss': round(avg_loss, 2),
'profit_factor': round(profit_factor, 3),
'sharpe_ratio': round(sharpe, 3),
'max_drawdown_pct': round(max_dd * 100, 2),
'avg_hold_bars': round(avg_hold, 1),
'long_trades': len(long_trades),
'long_pnl': round(long_pnl, 2),
'long_win_rate': round(long_wins / len(long_trades) * 100, 1) if long_trades else 0,
'short_trades': len(short_trades),
'short_pnl': round(short_pnl, 2),
'short_win_rate': round(short_wins / len(short_trades) * 100, 1) if short_trades else 0,
'exit_reasons': reason_stats,
'trades': trades[-20:], # Last 20 trades for inspection
}
# Print summary
logger.info(f"\n{'='*60}")
logger.info(f"BACKTEST RESULTS: {symbol}")
logger.info(f"{'='*60}")
logger.info(f" Status: {report['status']}")
logger.info(f" Total Trades: {report['total_trades']}")
logger.info(f" Total Return: {report['total_return_pct']}%")
logger.info(f" Total P&L: ${report['total_pnl']}")
logger.info(f" Win Rate: {report['win_rate_pct']}%")
logger.info(f" Avg Win: ${report['avg_win']}")
logger.info(f" Avg Loss: ${report['avg_loss']}")
logger.info(f" Profit Factor: {report['profit_factor']}")
logger.info(f" Sharpe Ratio: {report['sharpe_ratio']}")
logger.info(f" Max Drawdown: {report['max_drawdown_pct']}%")
logger.info(f" Avg Hold Time: {report['avg_hold_bars']} bars ({report['avg_hold_bars']:.0f}h)")
logger.info(f" LONG: {report['long_trades']} trades, ${report['long_pnl']} P&L, {report['long_win_rate']}% win")
logger.info(f" SHORT: {report['short_trades']} trades, ${report['short_pnl']} P&L, {report['short_win_rate']}% win")
logger.info(f"{'='*60}")
for reason, stats in reason_stats.items():
wr = stats['wins'] / stats['count'] * 100 if stats['count'] > 0 else 0
logger.info(f" {reason}: {stats['count']} exits, ${stats['total_pnl']:.2f} P&L, {wr:.0f}% win")
return report
def run_all_assets(self, days: int = 90) -> Dict:
"""Run backtest on all supported assets."""
assets = ['BTC/USDT', 'ETH/USDT', 'SOL/USDT', 'XRP/USDT']
results = {}
for asset in assets:
try:
results[asset] = self.run(asset, days)
except Exception as e:
logger.error(f"Backtest failed for {asset}: {e}")
results[asset] = {'error': str(e), 'status': 'ERROR'}
# Aggregate results
total_trades = sum(r.get('total_trades', 0) for r in results.values() if 'total_trades' in r)
total_pnl = sum(r.get('total_pnl', 0) for r in results.values() if 'total_pnl' in r)
total_wins = sum(r.get('wins', 0) for r in results.values() if 'wins' in r)
total_losses = sum(r.get('losses', 0) for r in results.values() if 'losses' in r)
agg_win_rate = total_wins / (total_wins + total_losses) * 100 if (total_wins + total_losses) > 0 else 0
avg_sharpe = np.mean([r.get('sharpe_ratio', 0) for r in results.values() if 'sharpe_ratio' in r])
max_dd = max([r.get('max_drawdown_pct', 0) for r in results.values() if 'max_drawdown_pct' in r], default=0)
passed = avg_sharpe > 0.5 and agg_win_rate > 50 and max_dd < 10
logger.info(f"\n{'='*60}")
logger.info(f"AGGREGATE BACKTEST RESULTS")
logger.info(f"{'='*60}")
logger.info(f" Overall Status: {'PASSED ✅' if passed else 'FAILED ❌'}")
logger.info(f" Total Trades: {total_trades}")
logger.info(f" Total P&L: ${total_pnl:.2f}")
logger.info(f" Win Rate: {agg_win_rate:.1f}%")
logger.info(f" Avg Sharpe: {avg_sharpe:.3f}")
logger.info(f" Max Drawdown: {max_dd:.2f}%")
logger.info(f"{'='*60}")
results['_aggregate'] = {
'status': 'PASSED ✅' if passed else 'FAILED ❌',
'total_trades': total_trades,
'total_pnl': round(total_pnl, 2),
'win_rate_pct': round(agg_win_rate, 1),
'avg_sharpe': round(avg_sharpe, 3),
'max_drawdown_pct': round(max_dd, 2),
}
# Save report
report_path = Path('./data/backtest_report.json')
report_path.parent.mkdir(parents=True, exist_ok=True)
# Make JSON serializable
serializable = {}
for k, v in results.items():
if isinstance(v, dict):
serializable[k] = {
sk: sv for sk, sv in v.items()
if not isinstance(sv, (np.floating, np.integer))
}
else:
serializable[k] = v
with open(report_path, 'w') as f:
json.dump(serializable, f, indent=2, default=str)
logger.info(f"\n📄 Report saved to {report_path}")
return results
def main():
parser = argparse.ArgumentParser(description='Strategy Backtester')
parser.add_argument('--asset', type=str, default=None, help='Single asset to backtest (e.g., BTCUSDT)')
parser.add_argument('--days', type=int, default=90, help='Days of history to backtest')
parser.add_argument('--balance', type=float, default=5000.0, help='Initial balance')
args = parser.parse_args()
backtester = StrategyBacktester(initial_balance=args.balance)
if args.asset:
symbol = args.asset.replace('USDT', '/USDT')
report = backtester.run(symbol, args.days)
else:
report = backtester.run_all_assets(args.days)
return report
if __name__ == '__main__':
main()