SciPy للحوسبة العلمية
أدوات علمية وإحصائية متقدمة. التحسين والتكامل والإحصاء ومعالجة الإشارات لأنظمة التداول.
الصعوبة: متوسط
الفئة: إحصائي
التثبيت
# Usually pre-installed with NumPy
$ pip install scipy
Key Modules for Trading
scipy.signal
Peak detection, filtering, and signal processing for price data
scipy.stats
Statistical distributions, hypothesis testing, and correlation analysis
scipy.optimize
Parameter optimization for trading strategies
scipy.interpolate
Fill gaps in price data and smooth noisy signals
scipy.fft
Fourier analysis to detect market cycles and periodicity
scipy.cluster
Hierarchical clustering for market regime detection
أمثلة الكود
Installation
Install SciPy
Python
pip install scipy
# Verify
python -c "import scipy; print(scipy.__version__)"
Support & Resistance Detection
Use find_peaks() to automatically detect support and resistance levels
Python
from scipy.signal import find_peaks
import numpy as np
import yfinance as yf
df = yf.download("EURUSD=X", period="6mo")
close = df['Close'].values.flatten()
high = df['High'].values.flatten()
low = df['Low'].values.flatten()
# Find resistance levels (peaks in price)
resistance_idx, resistance_props = find_peaks(
high, distance=10, prominence=0.001
)
# Find support levels (peaks in inverted price)
support_idx, support_props = find_peaks(
-low, distance=10, prominence=0.001
)
print("Resistance Levels:")
for idx in resistance_idx[-5:]:
print(f" {df.index[idx].date()}: {high[idx]:.5f}")
print("\nSupport Levels:")
for idx in support_idx[-5:]:
print(f" {df.index[idx].date()}: {low[idx]:.5f}")
Double Top/Bottom Detection
Identify double top and double bottom chart patterns
Python
from scipy.signal import find_peaks, argrelextrema
import numpy as np
import yfinance as yf
df = yf.download("EURUSD=X", period="1y")
close = df['Close'].values.flatten()
# Find local maxima (potential double tops)
peaks, _ = find_peaks(close, distance=15, prominence=0.0005)
# Check for double tops (two peaks at similar levels)
tolerance = 0.001 # 0.1% tolerance
double_tops = []
for i in range(len(peaks) - 1):
for j in range(i + 1, len(peaks)):
if abs(close[peaks[i]] - close[peaks[j]]) / close[peaks[i]] < tolerance:
if 10 < (peaks[j] - peaks[i]) < 60: # 10-60 bars apart
double_tops.append((peaks[i], peaks[j]))
print(f"Found {len(double_tops)} potential double tops:")
for top1, top2 in double_tops[:5]:
print(f" Peak 1: {df.index[top1].date()} ({close[top1]:.5f})")
print(f" Peak 2: {df.index[top2].date()} ({close[top2]:.5f})")
Noise Filtering with Savitzky-Golay
Smooth price data to reveal underlying trends
Python
from scipy.signal import savgol_filter
import numpy as np
import yfinance as yf
df = yf.download("EURUSD=X", period="3mo")
close = df['Close'].values.flatten()
# Apply Savitzky-Golay filter (preserves peaks better than moving average)
smooth_price = savgol_filter(close, window_length=21, polyorder=3)
# Calculate trend direction
trend = np.gradient(smooth_price)
print(f"Current price: {close[-1]:.5f}")
print(f"Smoothed price: {smooth_price[-1]:.5f}")
print(f"Trend direction: {'UP' if trend[-1] > 0 else 'DOWN'}")
print(f"Trend strength: {abs(trend[-1]):.6f}")
Market Cycle Detection with FFT
Find dominant cycles in price data using Fourier analysis
Python
from scipy.fft import fft, fftfreq
import numpy as np
import yfinance as yf
df = yf.download("EURUSD=X", period="2y")
close = df['Close'].values.flatten()
# Detrend the data
detrended = close - np.polyval(np.polyfit(range(len(close)), close, 1), range(len(close)))
# Apply FFT
N = len(detrended)
yf_vals = fft(detrended)
xf = fftfreq(N, d=1) # d=1 for daily frequency
# Get dominant frequencies (ignore DC component)
magnitudes = np.abs(yf_vals[1:N//2])
frequencies = xf[1:N//2]
periods = 1 / frequencies
# Top 5 dominant cycles
top_indices = np.argsort(magnitudes)[-5:][::-1]
print("Dominant Market Cycles:")
for idx in top_indices:
print(f" Period: {periods[idx]:.0f} days, "
f"Strength: {magnitudes[idx]:.4f}")
Strategy Parameter Optimization
Use scipy.optimize to find optimal indicator parameters
Python
from scipy.optimize import minimize
import numpy as np
import yfinance as yf
df = yf.download("EURUSD=X", period="2y")
close = df['Close'].values.flatten()
def strategy_return(params):
fast = int(params[0])
slow = int(params[1])
if fast >= slow or fast < 2 or slow > 200:
return 0 # Invalid parameters
# Simple SMA crossover backtest
sma_fast = np.convolve(close, np.ones(fast)/fast, mode='valid')
sma_slow = np.convolve(close, np.ones(slow)/slow, mode='valid')
# Align lengths
min_len = min(len(sma_fast), len(sma_slow))
sma_fast = sma_fast[-min_len:]
sma_slow = sma_slow[-min_len:]
aligned_close = close[-min_len:]
# Calculate returns
signals = np.where(sma_fast > sma_slow, 1, -1)
returns = np.diff(aligned_close) / aligned_close[:-1]
strategy_returns = signals[:-1] * returns
return -np.sum(strategy_returns) # Negative for minimization
# Optimize
result = minimize(strategy_return, x0=[10, 50],
method='Nelder-Mead',
bounds=[(2, 50), (20, 200)])
print(f"Optimal Fast SMA: {int(result.x[0])}")
print(f"Optimal Slow SMA: {int(result.x[1])}")
print(f"Total Return: {-result.fun:.4%}")
أفضل الممارسات
find_peaks() for S/R
Use prominence and distance parameters to filter meaningful support/resistance levels
Savgol Filter
Savitzky-Golay filter preserves peaks better than simple moving averages
Avoid Overfitting
When optimizing parameters, always validate on out-of-sample data
FFT Limitations
Market cycles are not perfectly periodic — use FFT as a guide, not a trading signal