# https://bibot.tistory.com/63 # https://nonmeyet.tistory.com/entry/Python-TALib%EB%A5%BC-%ED%99%9C%EC%9A%A9%ED%95%9C-%EB%B9%84%ED%8A%B8%EC%BD%94%EC%9D%B8%EC%A3%BC%EA%B0%80%EA%B8%B0%EC%88%A0%EB%B6%84%EC%84%9D-%EB%B3%B4%EC%A1%B0%EC%A7%80%ED%91%9C-%EC%B6%94%EA%B0%80 # https://lunadaddy.tistory.com/122 # https://wikidocs.net/186885 import os from scipy.signal import savgol_filter import numpy as np np.seterr(divide='ignore', invalid='ignore') # https://www.lfd.uci.edu/~gohlke/pythonlibs/#ta-lib # https://lunadaddy.tistory.com/122 import talib import pandas as pd from datetime import datetime, timedelta from stock.analysis.JSDPattern import JSDPattern class JSDPattern_realtime (JSDPattern): def __init__(self, stockFileName=None): super().__init__(stockFileName) return def analyze(self, result): result["volume"] = [result["volume"][i] if 0 < result["volume"][i] else 1 for i in range(len(result["volume"]))] # 기본 캔들 정보 open_df = pd.DataFrame(result["open"]) close_df = pd.DataFrame(result["close"]) high_df = pd.DataFrame(result["high"]) low_df = pd.DataFrame(result["low"]) volume_df = pd.DataFrame(result["volume"]) # 중복 제거 ymd_df = pd.DataFrame(result["ymd"]) data_dup = pd.concat([ymd_df, open_df, close_df, high_df, low_df, volume_df], axis=1) data_dup.columns = ["ymd", "open", "close", "high", "low", "volume"] data_dup.index = pd.DatetimeIndex(result["ymd"]) data_dup_sorted = data_dup.sort_index(ascending=True) data_dup_sorted = data_dup_sorted.drop_duplicates() ymd_df = data_dup_sorted["ymd"] open_df = data_dup_sorted["open"] close_df = data_dup_sorted["close"] high_df = data_dup_sorted["high"] low_df = data_dup_sorted["low"] volume_df = data_dup_sorted["volume"] ymd = ymd_df.tolist() open = open_df.tolist() close = close_df.tolist() high = high_df.tolist() low = low_df.tolist() volume = volume_df.tolist() # ichimokuCloud df = pd.concat([ymd_df, open_df, close_df, high_df, low_df, volume_df], axis=1) column_names = ['DATE', 'open', 'close', 'high', 'low', 'volume'] df.columns = column_names c, b, l, s = 9, 26, 52, 26 # 1. 전환선 = (과거 9일 동안 최고가 + 최저가) / 2 # 당일을 포함한 9일 동안의 최고가와 최저가의 중간 값을 평균으로 나타낸다. changeLine = (df.high.rolling(c).max() + df.low.rolling(c).min()) / 2 # 2. 기준선 = 과거 26일 동안 최고가 + 최저가) / 2 # 당일을 포함한 26일 동안의 최고가와 최저가의 중간 값을 평균으로 나타낸다. baseLine = (df.high.rolling(b).max() + df.low.rolling(b).min()) / 2 # 3. 후행스팬 = 현재 close가격의 26일전 반영 laggingSpan = [df.close.values[i + s] for i in range(len(df.close) - s)] laggingSpan += [None for i in range(s)] laggingSpan = np.array(laggingSpan) # 4. 선행스팬 1 = ((기준선 + 전환선) / 2)를 26일 선행하여 배치 # 전환선과 기준선의 평균값을 구해 당일 포함 26일 앞으로 이동시킨 선 (중-단기 구간의 힘을 보여줌) tmp_leadingSpan1 = (changeLine + baseLine) / 2 """ S: 26일 선행시킴 """ leadingSpan1 = list(tmp_leadingSpan1.values) for i in range(b - 1): leadingSpan1.insert(0, None) """ E: 26일 선행시킴 """ # 5. 선행스팬 2 = ((최근 52일 동안 최고가 + 최저가) / 2)를 26일 선행하여 배치 # 당일을 포함한 52일 동안의 최고가와 최저가의 평균을 26일 앞으로 이동시킨 선 (장기으로 형성된 선이기 때문에 가장 느리게 변함) tmp_leadingSpan2 = (df.high.rolling(l).max() + df.low.rolling(l).min()) / 2 """ S: 52일 선행시킴 """ leadingSpan2 = list(tmp_leadingSpan2.values) for i in range(l - 1): leadingSpan2.insert(0, None) """ S: 52일 선행시킴 """ baseLine = baseLine.tolist() changeLine = changeLine.tolist() laggingSpan = list(laggingSpan) current_index = len(ymd) for i in range(51): if len(ymd) < len(leadingSpan2): ymd.append(ymd[-1] + timedelta(days=1)) if len(open) < len(leadingSpan2): open.append(None) if len(close) < len(leadingSpan2): close.append(None) if len(high) < len(leadingSpan2): high.append(None) if len(low) < len(leadingSpan2): low.append(None) if len(volume) < len(leadingSpan2): volume.append(None) if len(baseLine) < len(leadingSpan2): baseLine.append(None) if len(changeLine) < len(leadingSpan2): changeLine.append(None) if len(laggingSpan) < len(leadingSpan2): laggingSpan.append(None) for i in range(26): if len(leadingSpan1) < len(leadingSpan2): leadingSpan1.append(leadingSpan1[-1]) # 9일 신고가 new_high_9 = [0 for c in range(8)] + [1 if (changeLine[c-1] is not None and changeLine[c] is not None and changeLine[c-1] < changeLine[c]) and None not in close[c-8:c+1] and max(close[c-8:c]) < close[c] else 0 for c in range(8, len(close))] # 26일 신고가 new_high_26 = [0 for c in range(25)] + [1 if (baseLine[c-1] is not None and baseLine[c] is not None and baseLine[c-1] < baseLine[c]) and None not in close[c-8:c+1] and max(close[c-25:c]) < close[c] else 0 for c in range(25, len(close))] # 9일 신저가 new_low_9 = [0 for c in range(8)] + [1 if (changeLine[c-1] is not None and changeLine[c] is not None and changeLine[c-1] < changeLine[c]) and None not in close[c-8:c+1] and close[c-9] < min(close[c-8:c+1]) else 0 for c in range(8, len(close))] # 26일 신저가 new_low_26 = [0 for c in range(25)] + [1 if (baseLine[c-1] is not None and baseLine[c] is not None and baseLine[c-1] < baseLine[c]) and None not in close[c-8:c+1] and close[c-26] < min(close[c-25:c+1]) else 0 for c in range(25, len(close))] # 이동 평균 close_df = pd.DataFrame(close) avg5 = list(np.reshape(close_df.ewm(5).mean().values, -1)) avg10 = list(np.reshape(close_df.ewm(10).mean().values, -1)) avg20 = list(np.reshape(close_df.ewm(20).mean().values, -1)) avg60 = list(np.reshape(close_df.ewm(60).mean().values, -1)) avg90 = list(np.reshape(close_df.ewm(90).mean().values, -1)) avg120 = list(np.reshape(close_df.ewm(120).mean().values, -1)) avg240 = list(np.reshape(close_df.ewm(240).mean().values, -1)) avg360 = list(np.reshape(close_df.ewm(360).mean().values, -1)) avg480 = list(np.reshape(close_df.ewm(480).mean().values, -1)) np_high, np_low, np_close = np.array(high, dtype=np.float64), np.array(low, dtype=np.float64), np.array(close, dtype=np.float64) slowk_12_df, slowd_12_df = talib.STOCH(np_high, np_low, np_close, fastk_period=12, slowk_period=5, slowk_matype=0, slowd_period=5, slowd_matype=0) slowk_26_df, slowd_26_df = talib.STOCH(np_high, np_low, np_close, fastk_period=26, slowk_period=16, slowk_matype=0, slowd_period=16, slowd_matype=0) slowk_52_df, slowd_52_df = talib.STOCH(np_high, np_low, np_close, fastk_period=52, slowk_period=32, slowk_matype=0, slowd_period=32, slowd_matype=0) # 볼린저 밴드 n, t = 10, 2 max_10 = close_df.rolling(window=n).mean() stddev_10 = close_df.rolling(window=n).std() upper_10 = max_10 + (stddev_10 * t) # 상단 볼리저 밴드 lower_10 = max_10 - (stddev_10 * t) # 하단 볼리저 밴드 middle_10 = (upper_10 + lower_10) / 2 upper_10 = list(np.reshape(upper_10.values, -1)) lower_10 = list(np.reshape(lower_10.values, -1)) middle_10 = list(np.reshape(middle_10.values, -1)) n, t = 20, 2 max_20 = close_df.rolling(window=n).mean() stddev_20 = close_df.rolling(window=n).std() upper_20 = max_20 + (stddev_20 * t) # 상단 볼리저 밴드 lower_20 = max_20 - (stddev_20 * t) # 하단 볼리저 밴드 middle_20 = (upper_20 + lower_20) / 2 upper_20 = list(np.reshape(upper_20.values, -1)) lower_20 = list(np.reshape(lower_20.values, -1)) middle_20 = list(np.reshape(middle_20.values, -1)) duration = 360 laggingSpan_close_diff, laggingSpan_close_diff_rate = self.getDiff_Rate(laggingSpan, close, duration=duration) laggingSpan_avg60_diff, laggingSpan_avg60_diff_rate = self.getDiff_Rate(laggingSpan, avg60, duration=duration) leadingSpan1_leadingSpan2_diff, leadingSpan1_leadingSpan2_diff_rate = self.getDiff_Rate(leadingSpan1, leadingSpan2, duration=duration) df_list = [ pd.DataFrame(ymd), pd.DataFrame(open), pd.DataFrame(close), pd.DataFrame(high), pd.DataFrame(low), pd.DataFrame(volume), pd.DataFrame(changeLine), pd.DataFrame(baseLine), pd.DataFrame(laggingSpan), pd.DataFrame(leadingSpan1), pd.DataFrame(leadingSpan2), pd.DataFrame(laggingSpan_close_diff), pd.DataFrame(laggingSpan_avg60_diff), pd.DataFrame(leadingSpan1_leadingSpan2_diff), pd.DataFrame(laggingSpan_close_diff_rate), pd.DataFrame(laggingSpan_avg60_diff_rate), pd.DataFrame(leadingSpan1_leadingSpan2_diff_rate), pd.DataFrame(avg5), pd.DataFrame(avg10), pd.DataFrame(avg20), pd.DataFrame(avg60), pd.DataFrame(avg90), pd.DataFrame(avg120), pd.DataFrame(avg240), pd.DataFrame(avg360), pd.DataFrame(avg480), pd.DataFrame(upper_10), pd.DataFrame(lower_10), pd.DataFrame(middle_10), pd.DataFrame(upper_20), pd.DataFrame(lower_20), pd.DataFrame(middle_20), pd.DataFrame(new_high_9), pd.DataFrame(new_high_26), pd.DataFrame(new_low_9), pd.DataFrame(new_low_26), pd.DataFrame(slowk_12_df), pd.DataFrame(slowd_12_df), pd.DataFrame(slowk_26_df), pd.DataFrame(slowd_26_df), pd.DataFrame(slowk_52_df), pd.DataFrame(slowd_52_df), ] data = pd.concat(df_list, axis=1) column_names = [ 'ymd', 'open', 'close', 'high', 'low', 'volume', 'changeLine', 'baseLine', 'laggingSpan', 'leadingSpan1', 'leadingSpan2', 'laggingSpan_close_diff', 'laggingSpan_avg60_diff', 'leadingSpan1_leadingSpan2_diff', 'laggingSpan_close_diff_rate', 'laggingSpan_avg60_diff_rate', 'leadingSpan1_leadingSpan2_diff_rate', 'avg5', 'avg10', 'avg20', 'avg60', 'avg90', 'avg120', 'avg240', 'avg360', 'avg480', 'upper_10', 'lower_10', 'middle_10', 'upper_20', 'lower_20', 'middle_20', 'new_high_9', 'new_high_26', 'new_low_9', 'new_low_26', 'slowk_12', 'slowd_12', 'slowk_26', 'slowd_26', 'slowk_52', 'slowd_52', ] data.columns = column_names data.index = pd.DatetimeIndex(ymd) return data, current_index def getData(self, ticker, ymd=None, get_days=14): if ymd is None: result = self.getCoinData(ticker, get_days=get_days) else: result = self.getCoinData(ticker, ymd=ymd, get_days=get_days) if len(result['ymd']) < 1: return None, None #result_tic = self.makeTickData(result_m1, mins=minute) data, current_index = self.analyze(result) return data, current_index if __name__ == "__main__": def min_max_normalize(data): min_val = min(data) max_val = max(data) normalized_data = [(x - min_val) / (max_val - min_val) for x in data] return normalized_data # 예시 데이터 original_data = [-4, -3, -2, -1, 0] normalized_data = min_max_normalize(original_data) print(np.asarray(normalized_data)-1) original_data = [0, 2,4,6,8,10] normalized_data = min_max_normalize(original_data) print(normalized_data)