Spaces:
Paused
Paused
Update ml_engine/monte_carlo.py
Browse files- ml_engine/monte_carlo.py +49 -64
ml_engine/monte_carlo.py
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
# ml_engine/monte_carlo.py
|
| 2 |
-
# (V11.
|
| 3 |
-
# - Tier 1: Light Check ->
|
| 4 |
-
# - Tier 2: Advanced Simulation ->
|
| 5 |
|
| 6 |
import numpy as np
|
| 7 |
import pandas as pd
|
|
@@ -13,37 +13,37 @@ class MonteCarloEngine:
|
|
| 13 |
"""
|
| 14 |
Advanced Stochastic Simulation Engine.
|
| 15 |
Provides two tiers of analysis:
|
| 16 |
-
1. Light Check: For rapid screening (
|
| 17 |
-
2. Advanced Simulation: For deep risk assessment (
|
| 18 |
"""
|
| 19 |
|
| 20 |
def __init__(self):
|
| 21 |
-
print("✅ [MonteCarlo V11.
|
| 22 |
|
| 23 |
# ==============================================================================
|
| 24 |
-
# ⚡ TIER 1: Light Check (
|
| 25 |
# ==============================================================================
|
| 26 |
def run_light_check(self, prices: List[float]) -> float:
|
| 27 |
"""
|
| 28 |
-
فحص سريع
|
| 29 |
-
|
|
|
|
|
|
|
|
|
|
| 30 |
|
| 31 |
Logic:
|
| 32 |
-
- Prob >
|
| 33 |
-
-
|
| 34 |
-
-
|
| 35 |
-
-
|
| 36 |
-
- Prob <
|
| 37 |
-
|
| 38 |
-
Returns:
|
| 39 |
-
float: Score between -0.10 and +0.10
|
| 40 |
"""
|
| 41 |
try:
|
| 42 |
-
# 1. تنظيف البيانات
|
| 43 |
-
if not prices or len(prices) <
|
| 44 |
return 0.0
|
| 45 |
|
| 46 |
-
# [CRITICAL] تحويل القائمة إلى float صراحة لتجنب مشاكل النصوص
|
| 47 |
clean_prices = []
|
| 48 |
for p in prices:
|
| 49 |
try:
|
|
@@ -51,63 +51,59 @@ class MonteCarloEngine:
|
|
| 51 |
if val > 0: clean_prices.append(val)
|
| 52 |
except: continue
|
| 53 |
|
| 54 |
-
if len(clean_prices) <
|
| 55 |
|
| 56 |
-
# 2. الحساب ال
|
| 57 |
price_arr = np.array(clean_prices, dtype=np.float64)
|
| 58 |
-
|
| 59 |
-
# ln(P_t / P_{t-1})
|
| 60 |
log_returns = np.diff(np.log(price_arr))
|
| 61 |
|
| 62 |
mu = np.mean(log_returns)
|
| 63 |
sigma = np.std(log_returns)
|
| 64 |
|
| 65 |
-
# إذا كان الانحراف صفراً (سعر ثابت)، العائد صفر
|
| 66 |
if sigma == 0: return 0.0
|
| 67 |
|
| 68 |
-
#
|
| 69 |
-
# Z-Score Approach: P(X > 0)
|
| 70 |
prob_up = norm.cdf(mu / sigma)
|
| 71 |
|
| 72 |
-
#
|
| 73 |
-
|
| 74 |
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
score = -0.05
|
| 83 |
-
else:
|
| 84 |
-
score = -0.10 # احتمالية الصعود ضعيفة جداً (< 40%)
|
| 85 |
-
|
| 86 |
-
# [DEBUG LOG] طباعة الاحتمالية للمراقبة (اختياري)
|
| 87 |
-
# print(f" 🎲 [Light MC] Prob: {prob_up:.2f} -> Score: {score}")
|
| 88 |
|
| 89 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 90 |
|
| 91 |
except Exception as e:
|
| 92 |
# print(f"⚠️ [Light MC] Error: {e}")
|
| 93 |
return 0.0
|
| 94 |
|
| 95 |
# ==============================================================================
|
| 96 |
-
# 🧠 TIER 2: Advanced Simulation (
|
| 97 |
# ==============================================================================
|
| 98 |
def run_advanced_simulation(self, prices: List[float], num_simulations=5000, time_horizon=24) -> float:
|
| 99 |
"""
|
| 100 |
-
المحاكاة المتقدمة
|
|
|
|
| 101 |
"""
|
| 102 |
try:
|
| 103 |
-
if not prices or len(prices) < 50:
|
| 104 |
-
return 0.0
|
| 105 |
-
|
| 106 |
-
# تنظيف البيانات
|
| 107 |
clean_prices = [float(p) for p in prices if p is not None]
|
| 108 |
if len(clean_prices) < 50: return 0.0
|
| 109 |
|
| 110 |
-
# 1. إحصائيات
|
| 111 |
closing_prices = np.array(clean_prices, dtype=np.float64)
|
| 112 |
log_returns = np.diff(np.log(closing_prices))
|
| 113 |
|
|
@@ -115,46 +111,35 @@ class MonteCarloEngine:
|
|
| 115 |
daily_volatility = np.std(log_returns)
|
| 116 |
last_price = closing_prices[-1]
|
| 117 |
|
| 118 |
-
#
|
| 119 |
random_shocks = np.random.normal(0, 1, (num_simulations, time_horizon))
|
| 120 |
drift = (mean_daily_return - 0.5 * daily_volatility ** 2)
|
| 121 |
diffusion = daily_volatility * random_shocks
|
| 122 |
-
|
| 123 |
daily_returns_sim = np.exp(drift + diffusion)
|
| 124 |
|
| 125 |
-
# مسارات الأسعار
|
| 126 |
price_paths = np.zeros_like(daily_returns_sim)
|
| 127 |
price_paths[:, 0] = last_price * daily_returns_sim[:, 0]
|
| 128 |
-
|
| 129 |
for t in range(1, time_horizon):
|
| 130 |
price_paths[:, t] = price_paths[:, t-1] * daily_returns_sim[:, t]
|
| 131 |
|
| 132 |
-
# 3. النتائج
|
| 133 |
final_prices = price_paths[:, -1]
|
| 134 |
|
| 135 |
-
#
|
| 136 |
threshold = last_price * 1.002
|
| 137 |
prob_profit = np.mean(final_prices > threshold)
|
| 138 |
var_95 = np.percentile(final_prices, 5)
|
| 139 |
drawdown_risk = (var_95 - last_price) / last_price
|
| 140 |
expected_return = (np.mean(final_prices) - last_price) / last_price
|
| 141 |
|
| 142 |
-
#
|
| 143 |
-
|
| 144 |
-
# 4. التنقيط (محدث ليكون أكثر توازناً)
|
| 145 |
mc_score = 0.0
|
| 146 |
-
|
| 147 |
-
# المكافآت
|
| 148 |
if prob_profit > 0.55: mc_score += 0.05
|
| 149 |
if prob_profit > 0.65: mc_score += 0.05
|
| 150 |
if expected_return > 0.005: mc_score += 0.05
|
| 151 |
-
|
| 152 |
-
# العقوبات
|
| 153 |
if drawdown_risk < -0.07: mc_score -= 0.05
|
| 154 |
if drawdown_risk < -0.10: mc_score -= 0.05
|
| 155 |
|
| 156 |
return round(max(-0.10, min(0.20, mc_score)), 2)
|
| 157 |
|
| 158 |
except Exception as e:
|
| 159 |
-
print(f"⚠️ [MonteCarlo] Advanced failed: {e}")
|
| 160 |
return 0.0
|
|
|
|
| 1 |
# ml_engine/monte_carlo.py
|
| 2 |
+
# (V11.7 - GEM-Architect: Fixed Light MC Sensitivity)
|
| 3 |
+
# - Tier 1: Light Check -> Continuous Score (No Dead Zones).
|
| 4 |
+
# - Tier 2: Advanced Simulation -> Standard GBM Logic (Top 10 Only).
|
| 5 |
|
| 6 |
import numpy as np
|
| 7 |
import pandas as pd
|
|
|
|
| 13 |
"""
|
| 14 |
Advanced Stochastic Simulation Engine.
|
| 15 |
Provides two tiers of analysis:
|
| 16 |
+
1. Light Check: For rapid screening (All 150 Coins) -> Returns sensitive micro-scores.
|
| 17 |
+
2. Advanced Simulation: For deep risk assessment (Top 10 Only).
|
| 18 |
"""
|
| 19 |
|
| 20 |
def __init__(self):
|
| 21 |
+
print("✅ [MonteCarlo V11.7] Hybrid Engine Loaded (Corrected Sensitivity).")
|
| 22 |
|
| 23 |
# ==============================================================================
|
| 24 |
+
# ⚡ TIER 1: Light Check (Applied to ALL 150 Coins)
|
| 25 |
# ==============================================================================
|
| 26 |
def run_light_check(self, prices: List[float]) -> float:
|
| 27 |
"""
|
| 28 |
+
فحص سريع يطبق على الـ 150 عملة.
|
| 29 |
+
|
| 30 |
+
[التعديل المطلوب]:
|
| 31 |
+
إلغاء المنطقة الميتة تماماً.
|
| 32 |
+
أي انحراف عن 0.50 يعطي نتيجة فوراً.
|
| 33 |
|
| 34 |
Logic:
|
| 35 |
+
- Prob > 0.55 -> +0.10
|
| 36 |
+
- Prob > 0.52 -> +0.07
|
| 37 |
+
- Prob > 0.50 -> +0.03 (أي نبض إيجابي يعطي درجة)
|
| 38 |
+
- Prob < 0.50 -> -0.03 (أي نبض سلبي يخصم درجة)
|
| 39 |
+
- Prob < 0.48 -> -0.07
|
| 40 |
+
- Prob < 0.45 -> -0.10
|
|
|
|
|
|
|
| 41 |
"""
|
| 42 |
try:
|
| 43 |
+
# 1. تنظيف البيانات
|
| 44 |
+
if not prices or len(prices) < 20:
|
| 45 |
return 0.0
|
| 46 |
|
|
|
|
| 47 |
clean_prices = []
|
| 48 |
for p in prices:
|
| 49 |
try:
|
|
|
|
| 51 |
if val > 0: clean_prices.append(val)
|
| 52 |
except: continue
|
| 53 |
|
| 54 |
+
if len(clean_prices) < 20: return 0.0
|
| 55 |
|
| 56 |
+
# 2. الحساب الإحصائي (Z-Score)
|
| 57 |
price_arr = np.array(clean_prices, dtype=np.float64)
|
|
|
|
|
|
|
| 58 |
log_returns = np.diff(np.log(price_arr))
|
| 59 |
|
| 60 |
mu = np.mean(log_returns)
|
| 61 |
sigma = np.std(log_returns)
|
| 62 |
|
|
|
|
| 63 |
if sigma == 0: return 0.0
|
| 64 |
|
| 65 |
+
# احتمالية الصعود P(X > 0)
|
|
|
|
| 66 |
prob_up = norm.cdf(mu / sigma)
|
| 67 |
|
| 68 |
+
# 3. سلم الحساسية (بدون أصفار - No Dead Zone)
|
| 69 |
+
# الهدف: التخلص من الـ 0.00 في الجدول
|
| 70 |
|
| 71 |
+
# --- المنطقة الإيجابية ---
|
| 72 |
+
if prob_up >= 0.55:
|
| 73 |
+
return 0.10 # صعود قوي
|
| 74 |
+
elif prob_up >= 0.52:
|
| 75 |
+
return 0.07 # صعود متوسط
|
| 76 |
+
elif prob_up > 0.5001:
|
| 77 |
+
return 0.03 # صعود طفيف جداً (لكي لا يكون صفر)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 78 |
|
| 79 |
+
# --- المنطقة السلبية ---
|
| 80 |
+
elif prob_up <= 0.45:
|
| 81 |
+
return -0.10 # هبوط قوي
|
| 82 |
+
elif prob_up <= 0.48:
|
| 83 |
+
return -0.07 # هبوط متوسط
|
| 84 |
+
elif prob_up < 0.4999:
|
| 85 |
+
return -0.03 # هبوط طفيف جداً (لكي لا يكون صفر)
|
| 86 |
+
|
| 87 |
+
# الحالة النادرة جداً (التعادل التام)
|
| 88 |
+
return 0.00
|
| 89 |
|
| 90 |
except Exception as e:
|
| 91 |
# print(f"⚠️ [Light MC] Error: {e}")
|
| 92 |
return 0.0
|
| 93 |
|
| 94 |
# ==============================================================================
|
| 95 |
+
# 🧠 TIER 2: Advanced Simulation (Applied to Top 10 Only)
|
| 96 |
# ==============================================================================
|
| 97 |
def run_advanced_simulation(self, prices: List[float], num_simulations=5000, time_horizon=24) -> float:
|
| 98 |
"""
|
| 99 |
+
المحاكاة المتقدمة للعملات المختارة فقط.
|
| 100 |
+
تعتمد منطق المخاطرة والعائد (Sharpe/VaR).
|
| 101 |
"""
|
| 102 |
try:
|
| 103 |
+
if not prices or len(prices) < 50: return 0.0
|
|
|
|
|
|
|
|
|
|
| 104 |
clean_prices = [float(p) for p in prices if p is not None]
|
| 105 |
if len(clean_prices) < 50: return 0.0
|
| 106 |
|
|
|
|
| 107 |
closing_prices = np.array(clean_prices, dtype=np.float64)
|
| 108 |
log_returns = np.diff(np.log(closing_prices))
|
| 109 |
|
|
|
|
| 111 |
daily_volatility = np.std(log_returns)
|
| 112 |
last_price = closing_prices[-1]
|
| 113 |
|
| 114 |
+
# GBM Simulation
|
| 115 |
random_shocks = np.random.normal(0, 1, (num_simulations, time_horizon))
|
| 116 |
drift = (mean_daily_return - 0.5 * daily_volatility ** 2)
|
| 117 |
diffusion = daily_volatility * random_shocks
|
|
|
|
| 118 |
daily_returns_sim = np.exp(drift + diffusion)
|
| 119 |
|
|
|
|
| 120 |
price_paths = np.zeros_like(daily_returns_sim)
|
| 121 |
price_paths[:, 0] = last_price * daily_returns_sim[:, 0]
|
|
|
|
| 122 |
for t in range(1, time_horizon):
|
| 123 |
price_paths[:, t] = price_paths[:, t-1] * daily_returns_sim[:, t]
|
| 124 |
|
|
|
|
| 125 |
final_prices = price_paths[:, -1]
|
| 126 |
|
| 127 |
+
# Metrics
|
| 128 |
threshold = last_price * 1.002
|
| 129 |
prob_profit = np.mean(final_prices > threshold)
|
| 130 |
var_95 = np.percentile(final_prices, 5)
|
| 131 |
drawdown_risk = (var_95 - last_price) / last_price
|
| 132 |
expected_return = (np.mean(final_prices) - last_price) / last_price
|
| 133 |
|
| 134 |
+
# Scoring (Standard Logic for Advanced)
|
|
|
|
|
|
|
| 135 |
mc_score = 0.0
|
|
|
|
|
|
|
| 136 |
if prob_profit > 0.55: mc_score += 0.05
|
| 137 |
if prob_profit > 0.65: mc_score += 0.05
|
| 138 |
if expected_return > 0.005: mc_score += 0.05
|
|
|
|
|
|
|
| 139 |
if drawdown_risk < -0.07: mc_score -= 0.05
|
| 140 |
if drawdown_risk < -0.10: mc_score -= 0.05
|
| 141 |
|
| 142 |
return round(max(-0.10, min(0.20, mc_score)), 2)
|
| 143 |
|
| 144 |
except Exception as e:
|
|
|
|
| 145 |
return 0.0
|