Spaces:
Running
Running
Create titan_engine.py
Browse files- ml_engine/titan_engine.py +152 -0
ml_engine/titan_engine.py
ADDED
|
@@ -0,0 +1,152 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# ml_engine/titan_engine.py
|
| 2 |
+
# (V1.0 - Titan Inference Engine)
|
| 3 |
+
|
| 4 |
+
import os
|
| 5 |
+
import joblib
|
| 6 |
+
import numpy as np
|
| 7 |
+
import pandas as pd
|
| 8 |
+
import pandas_ta as ta
|
| 9 |
+
import xgboost as xgb
|
| 10 |
+
import json
|
| 11 |
+
|
| 12 |
+
class TitanEngine:
|
| 13 |
+
def __init__(self, model_dir="ml_models/layer2"):
|
| 14 |
+
self.model_path = os.path.join(model_dir, "Titan_XGB_V1.json")
|
| 15 |
+
self.features_path = os.path.join(model_dir, "Titan_Features.pkl")
|
| 16 |
+
self.model = None
|
| 17 |
+
self.feature_names = None
|
| 18 |
+
self.initialized = False
|
| 19 |
+
|
| 20 |
+
async def initialize(self):
|
| 21 |
+
"""تحميل النموذج وقائمة الميزات من القرص"""
|
| 22 |
+
print(f"🛡️ [Titan] جاري تهيئة المحرك من {self.model_path}...")
|
| 23 |
+
try:
|
| 24 |
+
if os.path.exists(self.model_path) and os.path.exists(self.features_path):
|
| 25 |
+
# تحميل نموذج XGBoost
|
| 26 |
+
self.model = xgb.Booster()
|
| 27 |
+
self.model.load_model(self.model_path)
|
| 28 |
+
|
| 29 |
+
# تحميل قائمة الميزات لضمان الترتيب الصحيح
|
| 30 |
+
self.feature_names = joblib.load(self.features_path)
|
| 31 |
+
|
| 32 |
+
self.initialized = True
|
| 33 |
+
print(f"✅ [Titan] تم التحميل بنجاح. جاهز بـ {len(self.feature_names)} ميزة.")
|
| 34 |
+
else:
|
| 35 |
+
print(f"❌ [Titan] ملفات النموذج مفقودة!")
|
| 36 |
+
except Exception as e:
|
| 37 |
+
print(f"❌ [Titan] خطأ فادح أثناء التهيئة: {e}")
|
| 38 |
+
|
| 39 |
+
def apply_inverted_pyramid(self, df, tf):
|
| 40 |
+
"""نفس منطق هندسة الميزات المستخدم في التدريب تماماً"""
|
| 41 |
+
df = df.copy().sort_values('timestamp').reset_index(drop=True)
|
| 42 |
+
# تعيين الفهرس للسهولة في pandas_ta
|
| 43 |
+
df = df.set_index(pd.DatetimeIndex(pd.to_datetime(df['timestamp'], unit='ms')))
|
| 44 |
+
|
| 45 |
+
# --- المستوى 1: دقيق (5m, 15m) ---
|
| 46 |
+
if tf in ['5m', '15m']:
|
| 47 |
+
df['RSI'] = ta.rsi(df['close'], length=14)
|
| 48 |
+
df['MACD'] = ta.macd(df['close']).iloc[:, 0]
|
| 49 |
+
df['MACD_h'] = ta.macd(df['close']).iloc[:, 1]
|
| 50 |
+
df['CCI'] = ta.cci(df['high'], df['low'], df['close'], length=20)
|
| 51 |
+
df['ADX'] = ta.adx(df['high'], df['low'], df['close'], length=14).iloc[:, 0]
|
| 52 |
+
for p in [9, 21, 50, 200]:
|
| 53 |
+
ema = ta.ema(df['close'], length=p)
|
| 54 |
+
df[f'EMA_{p}_dist'] = (df['close'] / ema) - 1
|
| 55 |
+
bb = ta.bbands(df['close'], length=20, std=2.0)
|
| 56 |
+
df['BB_w'] = (bb.iloc[:, 2] - bb.iloc[:, 0]) / bb.iloc[:, 1]
|
| 57 |
+
df['BB_p'] = (df['close'] - bb.iloc[:, 0]) / (bb.iloc[:, 2] - bb.iloc[:, 0])
|
| 58 |
+
df['MFI'] = ta.mfi(df['high'], df['low'], df['close'], df['volume'], length=14)
|
| 59 |
+
vwap = ta.vwap(df['high'], df['low'], df['close'], df['volume'])
|
| 60 |
+
df['VWAP_dist'] = (df['close'] / vwap) - 1
|
| 61 |
+
|
| 62 |
+
# --- المستوى 2: تكتيكي (1h, 4h) ---
|
| 63 |
+
elif tf in ['1h', '4h']:
|
| 64 |
+
df['RSI'] = ta.rsi(df['close'], length=14)
|
| 65 |
+
df['MACD_h'] = ta.macd(df['close']).iloc[:, 1]
|
| 66 |
+
df['EMA_50_dist'] = (df['close'] / ta.ema(df['close'], length=50)) - 1
|
| 67 |
+
df['EMA_200_dist'] = (df['close'] / ta.ema(df['close'], length=200)) - 1
|
| 68 |
+
df['ATR_pct'] = ta.atr(df['high'], df['low'], df['close'], length=14) / df['close']
|
| 69 |
+
|
| 70 |
+
# --- المستوى 3: استراتيجي (1d) ---
|
| 71 |
+
elif tf == '1d':
|
| 72 |
+
df['RSI'] = ta.rsi(df['close'], length=14)
|
| 73 |
+
df['EMA_200_dist'] = (df['close'] / ta.ema(df['close'], length=200)) - 1
|
| 74 |
+
adx = ta.adx(df['high'], df['low'], df['close'])
|
| 75 |
+
if adx is not None and not adx.empty:
|
| 76 |
+
df['Trend_Strong'] = np.where(adx.iloc[:, 0] > 25, 1, 0)
|
| 77 |
+
else:
|
| 78 |
+
df['Trend_Strong'] = 0
|
| 79 |
+
|
| 80 |
+
return df.reset_index(drop=True)
|
| 81 |
+
|
| 82 |
+
def predict(self, ohlcv_data: dict) -> dict:
|
| 83 |
+
"""
|
| 84 |
+
استقبال البيانات الخام (Dictionary of DataFrames/Lists)،
|
| 85 |
+
تجهيزها، ثم استدعاء النموذج للتنبؤ.
|
| 86 |
+
"""
|
| 87 |
+
if not self.initialized or not self.model:
|
| 88 |
+
return {'score': 0.0, 'error': 'Titan not initialized'}
|
| 89 |
+
|
| 90 |
+
try:
|
| 91 |
+
# 1. تجهيز البيانات لكل إطار
|
| 92 |
+
processed_tfs = {}
|
| 93 |
+
for tf, data in ohlcv_data.items():
|
| 94 |
+
if not data: continue
|
| 95 |
+
# تحويل القوائم إلى DataFrame إذا لزم الأمر
|
| 96 |
+
if isinstance(data, list):
|
| 97 |
+
df = pd.DataFrame(data, columns=['timestamp', 'open', 'high', 'low', 'close', 'volume'])
|
| 98 |
+
else:
|
| 99 |
+
df = data.copy()
|
| 100 |
+
|
| 101 |
+
# تطبيق المؤشرات حسب الإطار
|
| 102 |
+
df = self.apply_inverted_pyramid(df, tf)
|
| 103 |
+
processed_tfs[tf] = df
|
| 104 |
+
|
| 105 |
+
# 2. الدمج (Alignment) للحصول على آخر ��قطة (Latest Snapshot)
|
| 106 |
+
if '5m' not in processed_tfs:
|
| 107 |
+
return {'score': 0.0, 'error': 'Missing 5m base timeframe'}
|
| 108 |
+
|
| 109 |
+
# نأخذ آخر صف فقط من الـ 5m كأساس
|
| 110 |
+
latest_5m = processed_tfs['5m'].iloc[-1:].copy()
|
| 111 |
+
latest_ts = latest_5m['timestamp'].iloc[0]
|
| 112 |
+
|
| 113 |
+
base_row = latest_5m.add_prefix('5m_').rename(columns={'5m_timestamp': 'timestamp'})
|
| 114 |
+
|
| 115 |
+
# دمج باقي الأطر (نأخذ آخر شمعة أغلقت قبل أو مع شمعة الـ 5m الحالية)
|
| 116 |
+
for tf, df in processed_tfs.items():
|
| 117 |
+
if tf == '5m': continue
|
| 118 |
+
# العثور على الشمعة المناسبة زمنياً
|
| 119 |
+
relevant_row = df[df['timestamp'] <= latest_ts].iloc[-1:].copy()
|
| 120 |
+
if relevant_row.empty: continue
|
| 121 |
+
|
| 122 |
+
# تجهيز الأعمدة للدمج
|
| 123 |
+
cols = [c for c in relevant_row.columns if c not in ['timestamp','open','high','low','close','volume']]
|
| 124 |
+
for col in cols:
|
| 125 |
+
base_row[f"{tf}_{col}"] = relevant_row[col].values[0]
|
| 126 |
+
|
| 127 |
+
# 3. تجهيز شعاع الإدخال (Feature Vector)
|
| 128 |
+
# التأكد من وجود كل الميزات المطلوبة بالترتيب الصحيح
|
| 129 |
+
input_data = []
|
| 130 |
+
for feat in self.feature_names:
|
| 131 |
+
val = base_row.get(feat, np.nan)
|
| 132 |
+
# إذا كانت القيمة مصفوفة أو سلسلة بانداز، نأخذ القيمة الأولى
|
| 133 |
+
if isinstance(val, (pd.Series, np.ndarray)):
|
| 134 |
+
val = val.iloc[0] if len(val) > 0 else np.nan
|
| 135 |
+
input_data.append(val)
|
| 136 |
+
|
| 137 |
+
# 4. التنبؤ
|
| 138 |
+
# تحويل إلى DMatrix (تنسيق XGBoost السريع)
|
| 139 |
+
dtest = xgb.DMatrix([input_data], feature_names=self.feature_names)
|
| 140 |
+
prediction = self.model.predict(dtest)[0] # إرجاع الاحتمالية الأولى
|
| 141 |
+
|
| 142 |
+
return {
|
| 143 |
+
'score': float(prediction),
|
| 144 |
+
'timestamp': int(latest_ts),
|
| 145 |
+
'status': 'OK'
|
| 146 |
+
}
|
| 147 |
+
|
| 148 |
+
except Exception as e:
|
| 149 |
+
# print(f"⚠️ [Titan Error] {e}")
|
| 150 |
+
import traceback
|
| 151 |
+
traceback.print_exc()
|
| 152 |
+
return {'score': 0.0, 'error': str(e)}
|