価格の波に乗ろう！ SOLっていうコインで試す、ちょっと特別な取引方法

"""
Hilbert Transform Trading Signal Generator
ヒルベルト変換を使用してサイクルとトレンドを検出
"""
import pandas as pd
import numpy as np


def calculate_hilbert_signals(df: pd.DataFrame,
                             period: int = 7) -> pd.DataFrame:
    """
    Hilbert Transform戦略のシグナル生成
    
    Parameters:
    -----------
    df : pd.DataFrame
        OHLCVデータ
    period : int
        基本期間（デフォルト: 7）
    
    Returns:
    --------
    pd.DataFrame
        シグナルが追加されたDataFrame
    """
    df = df.copy()
    
    # Weighted Close Price
    df['wcp'] = (df['high'] + df['low'] + 2 * df['close']) / 4
    
    # Smooth (4期間WMA)
    df['smooth'] = (4 * df['wcp'] + 3 * df['wcp'].shift(1) + 
                    2 * df['wcp'].shift(2) + df['wcp'].shift(3)) / 10
    
    # Detrender (高周波成分の除去)
    df['detrender'] = (0.0962 * df['smooth'] + 0.5769 * df['smooth'].shift(2) -
                       0.5769 * df['smooth'].shift(4) - 0.0962 * df['smooth'].shift(6)) * 0.075
    df['detrender'] = df['detrender'].shift(3)
    
    # Compute InPhase and Quadrature components
    df['q1'] = (0.0962 * df['detrender'] + 0.5769 * df['detrender'].shift(2) -
                0.5769 * df['detrender'].shift(4) - 0.0962 * df['detrender'].shift(6)) * 0.075
    df['i1'] = df['detrender'].shift(3)
    
    # Advance Phase components
    df['ji'] = (0.0962 * df['i1'] + 0.5769 * df['i1'].shift(2) -
                0.5769 * df['i1'].shift(4) - 0.0962 * df['i1'].shift(6)) * 0.075
    df['jq'] = (0.0962 * df['q1'] + 0.5769 * df['q1'].shift(2) -
                0.5769 * df['q1'].shift(4) - 0.0962 * df['q1'].shift(6)) * 0.075
    
    # Phasor components
    df['i2'] = df['i1'] - df['jq']
    df['q2'] = df['q1'] + df['ji']
    
    # Smooth the components
    df['i2'] = 0.2 * df['i2'] + 0.8 * df['i2'].shift(1).fillna(0)
    df['q2'] = 0.2 * df['q2'] + 0.8 * df['q2'].shift(1).fillna(0)
    
    # Homodyne Discriminator
    df['re'] = df['i2'] * df['i2'].shift(1) + df['q2'] * df['q2'].shift(1)
    df['im'] = df['i2'] * df['q2'].shift(1) - df['q2'] * df['i2'].shift(1)
    
    # Smooth
    df['re'] = 0.2 * df['re'] + 0.8 * df['re'].shift(1).fillna(0)
    df['im'] = 0.2 * df['im'] + 0.8 * df['im'].shift(1).fillna(0)
    
    # Compute Period
    df['period'] = 2 * np.pi / (np.arctan2(df['im'], df['re'] + 0.0001) + 0.0001)
    df['period'] = df['period'].clip(period * 0.67, period * 1.5)
    df['period'] = 0.2 * df['period'] + 0.8 * df['period'].shift(1).fillna(period)
    
    # Compute Phase
    df['phase'] = np.arctan2(df['q1'], df['i1'] + 0.0001) * 180 / np.pi
    
    # Sine and Lead Sine
    df['sine'] = np.sin(df['phase'] * np.pi / 180)
    df['lead_sine'] = np.sin((df['phase'] + 45) * np.pi / 180)
    
    # シグナル生成
    df['sine_prev'] = df['sine'].shift(1)
    df['lead_sine_prev'] = df['lead_sine'].shift(1)
    df['is_buy'] = (
        (df['sine'] > df['lead_sine']) & 
        (df['sine_prev'] <= df['lead_sine_prev'])
    ) & df['sine'].notna()
    df['is_sell'] = (
        (df['sine'] < df['lead_sine']) & 
        (df['sine_prev'] >= df['lead_sine_prev'])
    ) & df['sine'].notna()
    
    # 不要カラム削除
    drop_cols = ['wcp', 'smooth', 'detrender', 'q1', 'i1', 'ji', 'jq', 'i2', 'q2',
                 're', 'im', 'phase', 'sine_prev', 'lead_sine_prev']
    df.drop(drop_cols, axis=1, inplace=True, errors='ignore')
    
    return df