Spaces:
Paused
Paused
Update backtest_engine.py
Browse files- backtest_engine.py +32 -55
backtest_engine.py
CHANGED
|
@@ -1,10 +1,9 @@
|
|
| 1 |
# ============================================================
|
| 2 |
-
# 🧪 backtest_engine.py (V51.
|
| 3 |
# ============================================================
|
| 4 |
# التحديثات:
|
| 5 |
-
# 1. إ
|
| 6 |
-
# 2.
|
| 7 |
-
# 3. دمج الفلتر الأولي (Scanner) كجزء من المعادلة.
|
| 8 |
# ============================================================
|
| 9 |
|
| 10 |
import asyncio
|
|
@@ -29,9 +28,8 @@ CACHE_DIR = "backtest_cache_grid"
|
|
| 29 |
class MassiveOptimizer:
|
| 30 |
def __init__(self, data_manager):
|
| 31 |
self.dm = data_manager
|
| 32 |
-
#
|
| 33 |
-
|
| 34 |
-
self.GRID_DENSITY = 10
|
| 35 |
|
| 36 |
self.TARGET_COINS = [
|
| 37 |
'BTC/USDT', 'ETH/USDT', 'SOL/USDT', 'BNB/USDT', 'XRP/USDT',
|
|
@@ -40,36 +38,38 @@ class MassiveOptimizer:
|
|
| 40 |
]
|
| 41 |
|
| 42 |
if not os.path.exists(CACHE_DIR): os.makedirs(CACHE_DIR)
|
| 43 |
-
print(f"🧪 [Backtest Engine V51.
|
| 44 |
|
| 45 |
async def fetch_deep_history(self):
|
| 46 |
"""تحميل البيانات وتجهيزها للمعالجة السريعة"""
|
| 47 |
print(f"\n⏳ [Data] Pre-fetching history for Grid Search...")
|
| 48 |
end_time_ms = int(time.time() * 1000)
|
| 49 |
-
start_time_ms = end_time_ms - (14 * 24 * 60 * 60 * 1000)
|
| 50 |
|
| 51 |
for sym in self.TARGET_COINS:
|
| 52 |
safe_sym = sym.replace('/', '_')
|
| 53 |
file_path = f"{CACHE_DIR}/{safe_sym}.pkl"
|
| 54 |
|
| 55 |
-
# إذا الملف موجود وحديث، لا نحمله مرة أخرى
|
| 56 |
if os.path.exists(file_path): continue
|
| 57 |
|
| 58 |
print(f" ⬇️ Downloading {sym}...", end="", flush=True)
|
| 59 |
try:
|
| 60 |
-
# سحب شمعات 15 دقيقة (أسرع وأدق للفلتر الجديد)
|
| 61 |
candles = await self.dm.exchange.fetch_ohlcv(sym, '15m', since=start_time_ms, limit=1000)
|
| 62 |
if candles:
|
| 63 |
df = pd.DataFrame(candles, columns=['timestamp', 'open', 'high', 'low', 'close', 'volume'])
|
| 64 |
df = df.drop_duplicates(subset=['timestamp']).sort_values('timestamp')
|
| 65 |
for col in ['open', 'high', 'low', 'close', 'volume']: df[col] = df[col].astype(float)
|
| 66 |
|
| 67 |
-
# حساب المؤشرات مسبقاً (Vectorized
|
| 68 |
-
# 1. Titan Proxy (Trend)
|
| 69 |
df['ema50'] = df['close'].ewm(span=50).mean()
|
| 70 |
|
| 71 |
-
#
|
| 72 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 73 |
|
| 74 |
# BB
|
| 75 |
df['ma20'] = df['close'].rolling(20).mean()
|
|
@@ -90,7 +90,6 @@ class MassiveOptimizer:
|
|
| 90 |
def _worker_evaluate_batch(combinations_batch, market_data_files):
|
| 91 |
"""
|
| 92 |
يقوم هذا العامل بتقييم مجموعة من التوليفات (Batch) دفعة واحدة.
|
| 93 |
-
يعمل في Process منفصل تماماً = سرعة قصوى.
|
| 94 |
"""
|
| 95 |
results = []
|
| 96 |
|
|
@@ -110,37 +109,33 @@ class MassiveOptimizer:
|
|
| 110 |
|
| 111 |
for df in dfs:
|
| 112 |
# ---------------------------------------------------
|
| 113 |
-
# ⚡ Vectorized Signal Logic
|
| 114 |
# ---------------------------------------------------
|
| 115 |
-
# 1. Titan Score (Simulated)
|
| 116 |
-
# إذا السعر فوق المتوسط = 1.0، وإلا 0.2
|
| 117 |
titan_score = np.where(df['close'] > df['ema50'], 0.9, 0.3)
|
| 118 |
|
| 119 |
-
# 2. Scanner Score (Simulated)
|
| 120 |
-
|
| 121 |
-
|
| 122 |
-
bb_cond = np.where(df['close'] > df['bb_upper'], 1.0, 0.0) # اختراق
|
| 123 |
|
| 124 |
-
# دمج المؤشرات للكاشف
|
| 125 |
scanner_score = (rsi_cond * 0.7) + (bb_cond * 0.3)
|
| 126 |
|
| 127 |
# 3. Final Weighted Score
|
| 128 |
final_score = (titan_score * w_titan) + (scanner_score * w_scanner)
|
| 129 |
-
# Normalize (تقريباً)
|
| 130 |
final_score = final_score / (w_titan + w_scanner)
|
| 131 |
|
| 132 |
-
# 4. Generate Entries
|
| 133 |
signals = (final_score > entry_thresh)
|
| 134 |
|
| 135 |
# 5. Fast Loop for PnL
|
| 136 |
-
# (Looping over numpy array is fast enough here)
|
| 137 |
prices = df['close'].values
|
| 138 |
-
|
|
|
|
|
|
|
| 139 |
|
| 140 |
in_pos = False
|
| 141 |
entry_p = 0.0
|
| 142 |
|
| 143 |
-
# محاكاة سريعة
|
| 144 |
for i in range(len(prices)-1):
|
| 145 |
if not in_pos and sigs[i]:
|
| 146 |
in_pos = True
|
|
@@ -149,44 +144,36 @@ class MassiveOptimizer:
|
|
| 149 |
curr = prices[i]
|
| 150 |
pnl = (curr - entry_p) / entry_p
|
| 151 |
|
| 152 |
-
|
| 153 |
-
if pnl > 0.03 or pnl < -0.015: # TP 3%, SL 1.5%
|
| 154 |
total_pnl += pnl
|
| 155 |
total_trades += 1
|
| 156 |
in_pos = False
|
| 157 |
|
| 158 |
-
if total_trades > 5:
|
| 159 |
results.append({
|
| 160 |
'config': config,
|
| 161 |
'pnl': total_pnl,
|
| 162 |
'trades': total_trades,
|
| 163 |
-
'score': total_pnl * np.log(total_trades)
|
| 164 |
})
|
| 165 |
|
| 166 |
return results
|
| 167 |
|
| 168 |
# ==============================================================
|
| 169 |
-
# 🚀 The Grid Generator
|
| 170 |
# ==============================================================
|
| 171 |
async def run_optimization(self):
|
| 172 |
-
# 1. التأكد من البيانات
|
| 173 |
market_files = [os.path.join(CACHE_DIR, f) for f in os.listdir(CACHE_DIR) if f.endswith('.pkl')]
|
| 174 |
if not market_files:
|
| 175 |
await self.fetch_deep_history()
|
| 176 |
market_files = [os.path.join(CACHE_DIR, f) for f in os.listdir(CACHE_DIR) if f.endswith('.pkl')]
|
| 177 |
|
| 178 |
-
# 2. توليد الشبكة (The Grid)
|
| 179 |
print(f"🧩 [Optimizer] Generating Grid with Density={self.GRID_DENSITY}...")
|
| 180 |
|
| 181 |
-
# استخدام linspace لتوليد أرقام دقيقة بناءً على الكثافة
|
| 182 |
-
# كلما زاد self.GRID_DENSITY، زادت دقة الخطوات
|
| 183 |
w_titan_range = np.linspace(0.2, 0.9, num=self.GRID_DENSITY)
|
| 184 |
w_scanner_range = np.linspace(0.1, 0.8, num=self.GRID_DENSITY)
|
| 185 |
thresh_range = np.linspace(0.50, 0.80, num=self.GRID_DENSITY)
|
| 186 |
|
| 187 |
-
# يمكن إضافة المزيد من المتغيرات هنا لزيادة العدد لـ 100,000+
|
| 188 |
-
# مثلاً: scanner_rsi_limit = np.linspace(30, 70, num=5)
|
| 189 |
-
|
| 190 |
combinations = []
|
| 191 |
for wt, ws, th in itertools.product(w_titan_range, w_scanner_range, thresh_range):
|
| 192 |
combinations.append({
|
|
@@ -198,12 +185,9 @@ class MassiveOptimizer:
|
|
| 198 |
print(f" 📊 Total Unique Combinations: {len(combinations):,}")
|
| 199 |
print(f" 🚀 Est. Processing Time: {len(combinations)/2000:.1f} minutes (on parallel cores)")
|
| 200 |
|
| 201 |
-
# 3. التشغيل المتوازي (Multiprocessing)
|
| 202 |
start_time = time.time()
|
| 203 |
final_results = []
|
| 204 |
|
| 205 |
-
# تقسيم التوليفات إلى دفعات (Batches)
|
| 206 |
-
# كل نواة (Core) ستأخذ دفعة
|
| 207 |
batch_size = max(100, len(combinations) // (os.cpu_count() * 4))
|
| 208 |
batches = [combinations[i:i + batch_size] for i in range(0, len(combinations), batch_size)]
|
| 209 |
|
|
@@ -222,12 +206,10 @@ class MassiveOptimizer:
|
|
| 222 |
elapsed = time.time() - start_time
|
| 223 |
print(f"✅ Optimization Finished in {elapsed:.2f}s")
|
| 224 |
|
| 225 |
-
# 4. اختيار الفائز
|
| 226 |
if not final_results:
|
| 227 |
-
print("⚠️ No profitable strategies found.")
|
| 228 |
return None
|
| 229 |
|
| 230 |
-
# الترتيب حسب معادلة (الربح × ثبات الصفقات)
|
| 231 |
best_result = sorted(final_results, key=lambda x: x['score'], reverse=True)[0]
|
| 232 |
|
| 233 |
print("\n" + "="*60)
|
|
@@ -240,7 +222,7 @@ class MassiveOptimizer:
|
|
| 240 |
return best_result['config']
|
| 241 |
|
| 242 |
async def run_strategic_optimization_task():
|
| 243 |
-
print("\n🧪 [STRATEGIC BACKTEST V51.
|
| 244 |
from r2 import R2Service
|
| 245 |
r2 = R2Service()
|
| 246 |
dm = DataManager(None, None, r2)
|
|
@@ -253,19 +235,14 @@ async def run_strategic_optimization_task():
|
|
| 253 |
hub = AdaptiveHub(r2)
|
| 254 |
await hub.initialize()
|
| 255 |
|
| 256 |
-
|
| 257 |
-
regime = "RANGE" # أو اكتشافه
|
| 258 |
if regime in hub.strategies:
|
| 259 |
print(f"💉 Injecting new DNA into {regime} Strategy...")
|
| 260 |
st = hub.strategies[regime]
|
| 261 |
|
| 262 |
-
# تحديث الأوزان
|
| 263 |
st.model_weights['titan'] = best_config['w_titan']
|
| 264 |
-
|
| 265 |
-
# نفترض أننا نستخدم مفتاح 'patterns' لتخزين وزن الـ Scanner الجديد مؤقتاً
|
| 266 |
-
# أو نضيف حقلاً جديداً إذا عدلت الكلاس
|
| 267 |
st.model_weights['patterns'] = best_config['w_scanner']
|
| 268 |
-
|
| 269 |
st.filters['l1_min_score'] = best_config['thresh'] * 100
|
| 270 |
|
| 271 |
await hub._save_state_to_r2()
|
|
|
|
| 1 |
# ============================================================
|
| 2 |
+
# 🧪 backtest_engine.py (V51.1 - GEM-Architect: Bug Fix)
|
| 3 |
# ============================================================
|
| 4 |
# التحديثات:
|
| 5 |
+
# 1. إصلاح خطأ 'numpy.ndarray object has no attribute values'.
|
| 6 |
+
# 2. تحسين التعامل مع المصفوفات لضمان استقرار الباكتست.
|
|
|
|
| 7 |
# ============================================================
|
| 8 |
|
| 9 |
import asyncio
|
|
|
|
| 28 |
class MassiveOptimizer:
|
| 29 |
def __init__(self, data_manager):
|
| 30 |
self.dm = data_manager
|
| 31 |
+
# 3 = سريع (تجربة) | 5 = متوسط (~3000) | 10 = دقيق (~1000)
|
| 32 |
+
self.GRID_DENSITY = 10
|
|
|
|
| 33 |
|
| 34 |
self.TARGET_COINS = [
|
| 35 |
'BTC/USDT', 'ETH/USDT', 'SOL/USDT', 'BNB/USDT', 'XRP/USDT',
|
|
|
|
| 38 |
]
|
| 39 |
|
| 40 |
if not os.path.exists(CACHE_DIR): os.makedirs(CACHE_DIR)
|
| 41 |
+
print(f"🧪 [Backtest Engine V51.1] Grid Density set to: {self.GRID_DENSITY}")
|
| 42 |
|
| 43 |
async def fetch_deep_history(self):
|
| 44 |
"""تحميل البيانات وتجهيزها للمعالجة السريعة"""
|
| 45 |
print(f"\n⏳ [Data] Pre-fetching history for Grid Search...")
|
| 46 |
end_time_ms = int(time.time() * 1000)
|
| 47 |
+
start_time_ms = end_time_ms - (14 * 24 * 60 * 60 * 1000)
|
| 48 |
|
| 49 |
for sym in self.TARGET_COINS:
|
| 50 |
safe_sym = sym.replace('/', '_')
|
| 51 |
file_path = f"{CACHE_DIR}/{safe_sym}.pkl"
|
| 52 |
|
|
|
|
| 53 |
if os.path.exists(file_path): continue
|
| 54 |
|
| 55 |
print(f" ⬇️ Downloading {sym}...", end="", flush=True)
|
| 56 |
try:
|
|
|
|
| 57 |
candles = await self.dm.exchange.fetch_ohlcv(sym, '15m', since=start_time_ms, limit=1000)
|
| 58 |
if candles:
|
| 59 |
df = pd.DataFrame(candles, columns=['timestamp', 'open', 'high', 'low', 'close', 'volume'])
|
| 60 |
df = df.drop_duplicates(subset=['timestamp']).sort_values('timestamp')
|
| 61 |
for col in ['open', 'high', 'low', 'close', 'volume']: df[col] = df[col].astype(float)
|
| 62 |
|
| 63 |
+
# حساب المؤشرات مسبقاً (Vectorized)
|
|
|
|
| 64 |
df['ema50'] = df['close'].ewm(span=50).mean()
|
| 65 |
|
| 66 |
+
# Scanner Proxies
|
| 67 |
+
# RSI Manual Calculation for speed
|
| 68 |
+
delta = df['close'].diff()
|
| 69 |
+
gain = (delta.where(delta > 0, 0)).rolling(window=14).mean()
|
| 70 |
+
loss = (-delta.where(delta < 0, 0)).rolling(window=14).mean()
|
| 71 |
+
rs = gain / loss
|
| 72 |
+
df['rsi'] = 100 - (100 / (1 + rs))
|
| 73 |
|
| 74 |
# BB
|
| 75 |
df['ma20'] = df['close'].rolling(20).mean()
|
|
|
|
| 90 |
def _worker_evaluate_batch(combinations_batch, market_data_files):
|
| 91 |
"""
|
| 92 |
يقوم هذا العامل بتقييم مجموعة من التوليفات (Batch) دفعة واحدة.
|
|
|
|
| 93 |
"""
|
| 94 |
results = []
|
| 95 |
|
|
|
|
| 109 |
|
| 110 |
for df in dfs:
|
| 111 |
# ---------------------------------------------------
|
| 112 |
+
# ⚡ Vectorized Signal Logic
|
| 113 |
# ---------------------------------------------------
|
| 114 |
+
# 1. Titan Score (Simulated)
|
|
|
|
| 115 |
titan_score = np.where(df['close'] > df['ema50'], 0.9, 0.3)
|
| 116 |
|
| 117 |
+
# 2. Scanner Score (Simulated)
|
| 118 |
+
rsi_cond = np.where(df['rsi'] < 60, 1.0, 0.4)
|
| 119 |
+
bb_cond = np.where(df['close'] > df['bb_upper'], 1.0, 0.0)
|
|
|
|
| 120 |
|
|
|
|
| 121 |
scanner_score = (rsi_cond * 0.7) + (bb_cond * 0.3)
|
| 122 |
|
| 123 |
# 3. Final Weighted Score
|
| 124 |
final_score = (titan_score * w_titan) + (scanner_score * w_scanner)
|
|
|
|
| 125 |
final_score = final_score / (w_titan + w_scanner)
|
| 126 |
|
| 127 |
+
# 4. Generate Entries (Boolean Numpy Array)
|
| 128 |
signals = (final_score > entry_thresh)
|
| 129 |
|
| 130 |
# 5. Fast Loop for PnL
|
|
|
|
| 131 |
prices = df['close'].values
|
| 132 |
+
|
| 133 |
+
# 🔥 FIX: signals هو أصلاً numpy array، لا نحتاج .values
|
| 134 |
+
sigs = signals
|
| 135 |
|
| 136 |
in_pos = False
|
| 137 |
entry_p = 0.0
|
| 138 |
|
|
|
|
| 139 |
for i in range(len(prices)-1):
|
| 140 |
if not in_pos and sigs[i]:
|
| 141 |
in_pos = True
|
|
|
|
| 144 |
curr = prices[i]
|
| 145 |
pnl = (curr - entry_p) / entry_p
|
| 146 |
|
| 147 |
+
if pnl > 0.03 or pnl < -0.015:
|
|
|
|
| 148 |
total_pnl += pnl
|
| 149 |
total_trades += 1
|
| 150 |
in_pos = False
|
| 151 |
|
| 152 |
+
if total_trades > 5:
|
| 153 |
results.append({
|
| 154 |
'config': config,
|
| 155 |
'pnl': total_pnl,
|
| 156 |
'trades': total_trades,
|
| 157 |
+
'score': total_pnl * np.log(total_trades)
|
| 158 |
})
|
| 159 |
|
| 160 |
return results
|
| 161 |
|
| 162 |
# ==============================================================
|
| 163 |
+
# 🚀 The Grid Generator
|
| 164 |
# ==============================================================
|
| 165 |
async def run_optimization(self):
|
|
|
|
| 166 |
market_files = [os.path.join(CACHE_DIR, f) for f in os.listdir(CACHE_DIR) if f.endswith('.pkl')]
|
| 167 |
if not market_files:
|
| 168 |
await self.fetch_deep_history()
|
| 169 |
market_files = [os.path.join(CACHE_DIR, f) for f in os.listdir(CACHE_DIR) if f.endswith('.pkl')]
|
| 170 |
|
|
|
|
| 171 |
print(f"🧩 [Optimizer] Generating Grid with Density={self.GRID_DENSITY}...")
|
| 172 |
|
|
|
|
|
|
|
| 173 |
w_titan_range = np.linspace(0.2, 0.9, num=self.GRID_DENSITY)
|
| 174 |
w_scanner_range = np.linspace(0.1, 0.8, num=self.GRID_DENSITY)
|
| 175 |
thresh_range = np.linspace(0.50, 0.80, num=self.GRID_DENSITY)
|
| 176 |
|
|
|
|
|
|
|
|
|
|
| 177 |
combinations = []
|
| 178 |
for wt, ws, th in itertools.product(w_titan_range, w_scanner_range, thresh_range):
|
| 179 |
combinations.append({
|
|
|
|
| 185 |
print(f" 📊 Total Unique Combinations: {len(combinations):,}")
|
| 186 |
print(f" 🚀 Est. Processing Time: {len(combinations)/2000:.1f} minutes (on parallel cores)")
|
| 187 |
|
|
|
|
| 188 |
start_time = time.time()
|
| 189 |
final_results = []
|
| 190 |
|
|
|
|
|
|
|
| 191 |
batch_size = max(100, len(combinations) // (os.cpu_count() * 4))
|
| 192 |
batches = [combinations[i:i + batch_size] for i in range(0, len(combinations), batch_size)]
|
| 193 |
|
|
|
|
| 206 |
elapsed = time.time() - start_time
|
| 207 |
print(f"✅ Optimization Finished in {elapsed:.2f}s")
|
| 208 |
|
|
|
|
| 209 |
if not final_results:
|
| 210 |
+
print("⚠️ No profitable strategies found (Check Data or lowered thresholds).")
|
| 211 |
return None
|
| 212 |
|
|
|
|
| 213 |
best_result = sorted(final_results, key=lambda x: x['score'], reverse=True)[0]
|
| 214 |
|
| 215 |
print("\n" + "="*60)
|
|
|
|
| 222 |
return best_result['config']
|
| 223 |
|
| 224 |
async def run_strategic_optimization_task():
|
| 225 |
+
print("\n🧪 [STRATEGIC BACKTEST V51.1] Starting Massive Grid Search...")
|
| 226 |
from r2 import R2Service
|
| 227 |
r2 = R2Service()
|
| 228 |
dm = DataManager(None, None, r2)
|
|
|
|
| 235 |
hub = AdaptiveHub(r2)
|
| 236 |
await hub.initialize()
|
| 237 |
|
| 238 |
+
regime = "RANGE"
|
|
|
|
| 239 |
if regime in hub.strategies:
|
| 240 |
print(f"💉 Injecting new DNA into {regime} Strategy...")
|
| 241 |
st = hub.strategies[regime]
|
| 242 |
|
|
|
|
| 243 |
st.model_weights['titan'] = best_config['w_titan']
|
| 244 |
+
# نستخدم 'patterns' لحفظ وزن الـ Scanner مؤقتاً أو كما اتفقنا سابقاً
|
|
|
|
|
|
|
| 245 |
st.model_weights['patterns'] = best_config['w_scanner']
|
|
|
|
| 246 |
st.filters['l1_min_score'] = best_config['thresh'] * 100
|
| 247 |
|
| 248 |
await hub._save_state_to_r2()
|