optimized backtesting effiency

2026-04-11 17:23:26 +02:00
parent 766009b11a
commit 373e589297
10 changed files with 371780 additions and 330 deletions
@@ -0,0 +1,87 @@
+from data.loader import load_candles, resample_candles
+from engine.backtester import run_backtest
+from strategies.ict_strategy import ICTStrategy
+import time
+from tqdm import tqdm
+from joblib import Parallel, delayed
+
+print("Loading data...")
+candles_1m = load_candles("data/data1.csv")
+candles_5m = resample_candles(candles_1m, period=5)
+print(f"Loaded {len(candles_5m):,} 5-minute candles.\n")
+
+param_combos = []
+for session in ["london", "new_york"]:
+    for lookback in [3, 5, 7, 10]:
+        for ob_age in [20, 50, 80]:
+            for atr in [1.0, 1.5, 2.0, 2.5]:
+                for sweep in [True, False]:
+                    sweep_lbs = [5, 10, 15] if sweep else [0]
+                    for sweep_lb in sweep_lbs:
+                        param_combos.append({
+                            "session": session,
+                            "lookback": lookback,
+                            "ob_age": ob_age,
+                            "atr": atr,
+                            "sweep": sweep,
+                            "sweep_lb": sweep_lb
+                        })
+
+print(f"Starting parallel optimization of {len(param_combos)} combinations...\n")
+
+def run_one_combo(params):
+    strategy = ICTStrategy(
+        session=params["session"],
+        lookback=params["lookback"],
+        ob_max_age=params["ob_age"],
+        atr_mult=params["atr"],
+        use_liquidity_sweep=params["sweep"],
+        sweep_lookback=params["sweep_lb"],
+    )
+
+    t0 = time.perf_counter()
+    trades = run_backtest(candles_5m, strategy, 10000)
+    elapsed = time.perf_counter() - t0
+
+    if len(trades) < 5:
+        return None
+
+    total_pnl = sum(t.pnl for t in trades)
+    wr = len([t for t in trades if t.pnl > 0]) / len(trades) * 100 if trades else 0.0
+
+    return {
+        "params": params,
+        "pnl": total_pnl,
+        "trades": len(trades),
+        "wr": round(wr, 2),
+        "time": round(elapsed, 4)
+    }
+
+total_start = time.perf_counter()
+
+results = Parallel(n_jobs=-1, verbose=10)(
+    delayed(run_one_combo)(params) for params in param_combos
+)
+
+valid_results = [r for r in results if r is not None]
+
+if not valid_results:
+    print("No valid strategies found with at least 5 trades.")
+    exit()
+
+best_result = max(valid_results, key=lambda x: x["pnl"])
+
+total_time = time.perf_counter() - total_start
+
+print("\n" + "="*70)
+print("PARALLEL OPTIMIZATION FINISHED!")
+print(f"Total time on M4 Mac: {total_time:.1f} seconds ({total_time/60:.1f} minutes)")
+print(f"Processed {len(param_combos)} combinations at ~{len(param_combos)/total_time:.2f} combos/second")
+print(f"Best PnL: {best_result['pnl']:.2f}")
+print(f"Best Params: {best_result['params']}")
+print(f"Trades: {best_result['trades']} | Win Rate: {best_result['wr']}%")
+print("="*70)
+
+print("\nTop 5 results:")
+for res in sorted(valid_results, key=lambda x: x["pnl"], reverse=True)[:5]:
+    print(f"PnL: {res['pnl']:.2f} | Trades: {res['trades']} | WR: {res['wr']}% | {res['params']}")