def preproc(self):
self.dat = df = pd.read_csv(self.path)
s = np.asanyarray(ta.stoch(
df["High"], df["Low"], df["Close"], 14)).reshape(
(-1, 1)) - np.asanyarray(
ta.stoch_signal(df["High"], df["Low"], df["Close"],
14)).reshape((-1, 1))
m = np.asanyarray(ta.macd(df["Close"])).reshape(
(-1, 1)) - np.asanyarray(ta.macd_signal(df["Close"])).reshape(
(-1, 1))
trend3 = np.asanyarray(self.dat[["Close"]]) - np.asanyarray(
ta.ema(self.dat["Close"], 20)).reshape((-1, 1))
cross1 = np.asanyarray(ta.ema(self.dat["Close"], 20)).reshape(
(-1, 1)) - np.asanyarray(ta.ema(self.dat["Close"], 5)).reshape(
(-1, 1))
y = np.asanyarray(self.dat[["Open"]])
x = np.concatenate([s, m, cross1], 1)
gen = tf.keras.preprocessing.sequence.TimeseriesGenerator(
x, y, self.window_size)
self.x = []
self.y = []
for i in gen:
self.x.extend(i[0].tolist())
self.y.extend(i[1].tolist())
self.x = np.asanyarray(
self.x) #.reshape((-1, self.window_size, x.shape[-1]))
self.y = np.asanyarray(self.y)
self.df = self.x
self.trend = self.y
def stoch(candles, window):
"""Stochastic Oscillator"""
highs = util.filtership(candles, "max")
lows = util.filtership(candles, "min")
closes = util.filtership(candles, "close")
result = ta.stoch(highs, lows, closes, window)
return result
def _indicadores(self, df):
''' ESTA FUNCAO NAO E EXECUDA, ELA EXECUTA ATRAVES DA FUNCAO INDICADOR() '''
import warnings
warnings.filterwarnings("ignore", message="divide by zero encountered in double_scalars")
warnings.filterwarnings("ignore", message="invalid value encountered in double_scalars")
db = pd.DataFrame()
for i in df.asset.unique():
df1 = df[df['asset'] == i].copy()
df1['PD_low'] = df1.low.shift(1)
df1['PD_high'] = df1.high.shift(1)
df1['Change%'] = df1.close.pct_change().round(4)
df1['SMA20'] = df1.close.rolling(20).mean().round(5)
df1['SMA200'] = df1.close.rolling(200).mean().round(5)
df1['low60'] = df1.low.rolling(window=60).min()
df1['low300'] = df1.low.rolling(window=300).min()
df1['low1200'] = df1.low.rolling(window=1200).min()
df1['high60'] = df1.high.rolling(window=60).max()
df1['high300'] = df1.high.rolling(window=300).max()
df1['high1200'] = df1.high.rolling(window=1200).max()
df1['%60'] = ((df1.close - df1.low60) / (df1.high60 - df1.low60)).round(2)
df1['%300'] = ((df1.close - df1.low300) / (df1.high300 - df1.low300)).round(2)
df1['%1200'] = ((df1.close - df1.low1200) / (df1.high1200 - df1.low1200)).round(2)
df1['ATR'] = ((df1.high - df1.low).rolling(10).mean() * 5 +
(df1.high - df1.low).rolling(60).mean() * 3 +
(df1.high - df1.low).rolling(1200).mean() * 2) / 10
df1['PW_low1'] = df1.low.shift(1).resample('W').min().reindex(df1.index, method='ffill') #FIXED
df1['PW_open'] = df1.open.shift(1).resample('W').first().reindex(df1.index, method='ffill') #FIXED
df1['PW_close'] = df1.close.shift(1).resample('W').last().reindex(df1.index, method='ffill') #FIXED
df1['PW_high1'] = df1.high.shift(1).resample('W').max().reindex(df1.index, method='ffill') #FIXED
df1['PM_low1'] = df1.low.resample('M').min().reindex(df1.index, method='ffill').shift(1) #FIXED
df1['PM_open'] = df1.open.resample('M').first().reindex(df1.index, method='ffill').shift(1) #FIXED
df1['PM_close'] = df1.close.resample('M').last().reindex(df1.index, method='ffill').shift(1) #FIXED
df1['PM_high1'] = df1.high.resample('M').max().reindex(df1.index, method='ffill').shift(1) #FIXED
df1['M1_low'] = df1.low.resample('30D').min().reindex(df1.index, method='ffill')
df1['M1_high'] = df1.high.resample('30D').max().reindex(df1.index, method='ffill')
df1['RSI_5K'] = ta.stoch(df.high, df.low, df.close, 5).round(0) #FIXED
df1['RSI_5D'] = ta.stoch_signal(df.high, df.low, df.close, 5, 3).round(0) #FIXED
df1['d30%'] = ((df1.close - df1.close.shift(30)) / df1.close).round(4)
db = pd.concat([db, df1])
return db
def rsi(self, df, period=50): #UPGRADE
df.columns = map(str.lower, df.columns) #UPGRADE
df.sort_index(inplace=True)
df = df.reset_index() #UPGRADE
rsi_k = ta.stoch(df['high'], df['low'], df['close'], period)
rsi_k = rsi_k.iloc[-1]
rsi_d = ta.stoch_signal(df['high'], df['low'], df['close'], period, 3)
rsi_d = rsi_d.iloc[-1]
return rsi_d, rsi_k
def rsi(self, df, period=5):
df.columns = map(str.lower, df.columns)
df.sort_index(inplace = True)
df = df.reset_index()
rsi_k = ta.stoch(df['high'], df['low'], df['close'], period)
rsi_k = rsi_k.iloc[-1]
rsi_d = ta.stoch_signal(df['high'], df['low'], df['close'], period, 3)
rsi_d = rsi_d.iloc[-1]
return int(rsi_d), int(rsi_k)
def so():
so = ta.stoch(high, low, close, n=14, fillna=False)
SO = (so[-1])
if SO <= 20:
status3 = "Buy"
elif SO <= 30:
status3 = "Buy"
elif SO >= 80:
status3 = "Sell"
elif SO >= 70:
status3 = "Sell"
else:
status3 = "Hold"
return status3
def do_ta(self, data_series):
open = Series(data_series['open'].astype('float64'))
high = Series(data_series['high'].astype('float64'))
low = Series(data_series['low'].astype('float64'))
close = Series(data_series['close'].astype('float64'))
# Trend
# ----------------
ema30 = ta.ema(series=close, periods=30)
ema50 = ta.ema(series=close, periods=50)
ema100 = ta.ema(series=close, periods=100)
ema200 = ta.ema(series=close, periods=200)
macd_diff = ta.macd_diff(close=close, n_fast=12, n_slow=26, n_sign=9)
macd_signal = ta.macd_signal(close=close,
n_fast=12,
n_slow=26,
n_sign=9)
data_series['ema30'] = ema30
data_series['ema50'] = ema50
data_series['ema100'] = ema100
data_series['ema200'] = ema200
data_series['macd_diff'] = macd_diff
data_series['macd_signal'] = macd_signal
# Momentum
# ----------------
rsi = ta.rsi(close=close)
stochastic = ta.stoch(high=high, low=low, close=close)
data_series['rsi'] = rsi
data_series['stochastic'] = stochastic
# Volatility
# ----------------
bollinger_h = ta.bollinger_hband(close=close)
bollinger_l = ta.bollinger_lband(close=close)
bollinger_h_indicator = ta.bollinger_hband_indicator(close=close)
bollinger_l_indicator = ta.bollinger_lband_indicator(close=close)
data_series['bollinger_h'] = bollinger_h
data_series['bollinger_l'] = bollinger_l
data_series['bollinger_h_indicator'] = bollinger_h_indicator
data_series['bollinger_l_indicator'] = bollinger_l_indicator
data_series['last_candle_change'] = self.lcc(close=close)
return data_series
def analyze(data):
# Moving average (5, 20, 75 days)
sma5 = data['close'].rolling(window=5).mean()
sma20 = data['close'].rolling(window=20).mean()
sma75 = data['close'].rolling(window=75).mean()
# MACD (12, 26 days)
macd12_26 = ta.macd(data['close'], n_fast=12, n_slow=26)
macd12_26_sig = ta.macd_signal(data['close'],
n_fast=12,
n_slow=26,
n_sign=9)
# Stochastics (%K: 5(not used), 9(not used), 14 day, %D: 3 days)
stoch14_k = ta.stoch(data['high'], data['low'], data['close'], n=14)
stoch14_d = ta.stoch_signal(data['high'],
data['low'],
data['close'],
n=14,
d_n=3)
# Bollinger band (+3 sigma ~ -3 sigma)
bb20_h1 = ta.bollinger_hband(data['close'], n=20, ndev=1)
bb20_h2 = ta.bollinger_hband(data['close'], n=20, ndev=2)
bb20_h3 = ta.bollinger_hband(data['close'], n=20, ndev=3)
bb20_l1 = ta.bollinger_lband(data['close'], n=20, ndev=1)
bb20_l2 = ta.bollinger_lband(data['close'], n=20, ndev=2)
bb20_l3 = ta.bollinger_lband(data['close'], n=20, ndev=3)
# Concatenate
analysis = pd.concat([
sma5, sma20, sma75, macd12_26, macd12_26_sig, stoch14_k, stoch14_d,
bb20_h1, bb20_h2, bb20_h3, bb20_l1, bb20_l2, bb20_l3
],
axis=1)
analysis.columns = [
'5-Day SMA', '20-Day SMA', '75-Day SMA', 'MACD (12-16)',
'MACD Signal (12-26-9)', 'Stochastics %K (14 Day)',
'Stochastics %D (14 Day)', 'BB +1sigma', 'BB +2sigma', 'BB +3sigma',
'BB -1sigma', 'BB -2sigma', 'BB -3sigma'
]
return analysis
def preproc(self):
self.dat = df = pd.read_csv(self.path)
s = np.asanyarray(ta.stoch(df["High"],df["Low"],df["Close"],14)).reshape((-1, 1)) - np.asanyarray(ta.stoch_signal(df["High"],df["Low"],df["Close"],14)).reshape((-1, 1))
x = np.asanyarray(ta.daily_return(df["Close"])).reshape((-1,1))
m = np.asanyarray(ta.macd_diff(df["Close"])).reshape((-1,1))
cross1 = np.asanyarray(ta.ema(self.dat["Close"],20)).reshape((-1, 1)) - np.asanyarray(ta.ema(self.dat["Close"],5)).reshape((-1, 1))
x = np.concatenate([x], 1)
y = np.asanyarray(self.dat[["Open"]])
gen = tf.keras.preprocessing.sequence.TimeseriesGenerator(x, y, self.window_size)
self.x = []
self.y = []
for i in gen:
self.x.extend(i[0].tolist())
self.y.extend(i[1].tolist())
self.x = np.asanyarray(self.x)[1000::]
self.y = np.asanyarray(self.y)[1000::]
self.df = self.x[-self.STEP_SIZE::]
self.trend = self.y[-self.STEP_SIZE::]
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import ta
df = pd.read_csv('./data/coinbase_daily.csv')
df = df.dropna().reset_index().sort_values('Date')
ta_df = pd.DataFrame()
ta_df['RSI'] = ta.rsi(df["Close"])
ta_df['MFI'] = ta.money_flow_index(
df["High"], df["Low"], df["Close"], df["Volume BTC"])
ta_df['TSI'] = ta.tsi(df["Close"])
ta_df['UO'] = ta.uo(df["High"], df["Low"], df["Close"])
ta_df['Stoch'] = ta.stoch(df["High"], df["Low"], df["Close"])
ta_df['Stoch_Signal'] = ta.stoch_signal(df["High"], df["Low"], df["Close"])
ta_df['WR'] = ta.wr(df["High"], df["Low"], df["Close"])
ta_df['AO'] = ta.ao(df["High"], df["Low"])
ta_df['MACD'] = ta.macd(df["Close"])
ta_df['MACD_signal'] = ta.macd_signal(df["Close"])
ta_df['MACD_diff'] = ta.macd_diff(df["Close"])
ta_df['EMA_fast'] = ta.ema_indicator(df["Close"])
ta_df['EMA_slow'] = ta.ema_indicator(df["Close"])
ta_df['Vortex_pos'] = ta.vortex_indicator_pos(
df["High"], df["Low"], df["Close"])
ta_df['Vortex_neg'] = ta.vortex_indicator_neg(
df["High"], df["Low"], df["Close"])
ta_df['Vortex_diff'] = abs(
ta_df['Vortex_pos'] -
def get_data(context, data_, window):
# Crear ventana de datos.
h1 = data_.history(
context.symbols,
context.row_features,
bar_count=window,
frequency=str(context.bar_period) + "T",
)
h1 = h1.swapaxes(2, 0)
norm_data = []
close_prices = []
for i, asset in enumerate(context.assets):
data = h1.iloc[i]
close = h1.iloc[i].close
if context.include_ha:
ha = heikenashi(data)
data = pd.concat((data, ha), axis=1)
for period in [3, 6, 8, 10, 15, 20]:
data["rsi" + str(period)] = ta.rsi(data.close,
n=period,
fillna=True)
data["stoch" + str(period)] = ta.stoch(data.high,
data.low,
data.close,
n=period,
fillna=True)
data["stoch_signal" + str(period)] = ta.stoch_signal(
high=data.high,
low=data.low,
close=data.close,
n=period,
d_n=3,
fillna=True)
data["dpo" + str(period)] = ta.dpo(close=data.close,
n=period,
fillna=True)
data["atr" + str(period)] = ta.average_true_range(high=data.high,
low=data.low,
close=data.close,
n=period,
fillna=True)
for period in [6, 7, 8, 9, 10]:
data["williams" + str(period)] = ta.wr(high=data.high,
low=data.low,
close=data.close,
lbp=period,
fillna=True)
for period in [12, 13, 14, 15]:
data["proc" + str(period)] = ta.trix(close=data.close,
n=period,
fillna=True)
data["macd_diff"] = ta.macd_diff(close=data.close,
n_fast=15,
n_slow=30,
n_sign=9,
fillna=True)
data["macd_signal"] = ta.macd_signal(close=data.close,
n_fast=15,
n_slow=30,
n_sign=9,
fillna=True)
data["bb_high_indicator"] = ta.bollinger_hband_indicator(
close=data.close, n=15, ndev=2, fillna=True)
data["bb_low_indicator"] = ta.bollinger_lband_indicator(
close=data.close, n=15, ndev=2, fillna=True)
data["dc_high_indicator"] = ta.donchian_channel_hband_indicator(
close=data.close, n=20, fillna=True)
data["dc_low_indicator"] = ta.donchian_channel_lband_indicator(
close=data.close, n=20, fillna=True)
data["ichimoku_a"] = ta.ichimoku_a(high=data.high,
low=data.low,
n1=9,
n2=26,
fillna=True)
data.fillna(method="bfill")
# Normalizar los valores
for feature in data.columns:
norm_feature = preprocessing.normalize(
data[feature].values.reshape(-1, 1), axis=0)
data[feature] = pd.DataFrame(data=norm_feature,
index=data.index,
columns=[feature])
norm_data.append(data.values)
close_prices.append(close)
context.features = data.columns
return np.array(norm_data), np.array(close_prices)
def get_trayectory(self, t_intervals):
"""
:param t_intervals: número de intervalos en cada trayectoria
:return: Datos con características de la trayectoria sintética y precios de cierre en bruto de al misma
"""
trayectories = []
closes = []
p = True
for i, asset in enumerate(self.context.assets):
synthetic_return = np.exp(
self.drift[i] + self.stdev[i] * norm.ppf(np.random.rand((t_intervals * self.frequency) + self.frequency, 1)))
initial_close = self.close[i, -1]
synthetic_close = np.zeros_like(synthetic_return)
synthetic_close[0] = initial_close
for t in range(1, synthetic_return.shape[0]):
synthetic_close[t] = synthetic_close[t - 1] * synthetic_return[t]
OHLC = []
for t in range(synthetic_return.shape[0]):
if t % self.frequency == 0 and t > 0:
open = synthetic_close[t - self.frequency]
high = np.max(synthetic_close[t - self.frequency: t])
low = np.min(synthetic_close[t - self.frequency: t])
close = synthetic_close[t]
OHLC.append([open, high, close, low])
data = pd.DataFrame(data=OHLC, columns=["open", "high", "low", "close"])
close = data.close
if self.context.include_ha:
ha = heikenashi(data)
data = pd.concat((data, ha), axis=1)
for period in [3, 6, 8, 10, 15, 20]:
data["rsi" + str(period)] = ta.rsi(data.close, n=period, fillna=True)
data["stoch" + str(period)] = ta.stoch(data.high, data.low, data.close, n=period, fillna=True)
data["stoch_signal" + str(period)] = ta.stoch_signal(high=data.high,
low=data.low,
close=data.close,
n=period,
d_n=3,
fillna=True)
data["dpo" + str(period)] = ta.dpo(close=data.close,
n=period,
fillna=True)
data["atr" + str(period)] = ta.average_true_range(high=data.high,
low=data.low,
close=data.close,
n=period,
fillna=True)
for period in [6, 7, 8, 9, 10]:
data["williams" + str(period)] = ta.wr(high=data.high,
low=data.low,
close=data.close,
lbp=period,
fillna=True)
for period in [12, 13, 14, 15]:
data["proc" + str(period)] = ta.trix(close=data.close,
n=period,
fillna=True)
data["macd_diff"] = ta.macd_diff(close=data.close,
n_fast=15,
n_slow=30,
n_sign=9,
fillna=True)
data["macd_signal"] = ta.macd_signal(close=data.close,
n_fast=15,
n_slow=30,
n_sign=9,
fillna=True)
data["bb_high_indicator"] = ta.bollinger_hband_indicator(close=data.close,
n=15,
ndev=2,
fillna=True)
data["bb_low_indicator"] = ta.bollinger_lband_indicator(close=data.close,
n=15,
ndev=2,
fillna=True)
data["dc_high_indicator"] = ta.donchian_channel_hband_indicator(close=data.close,
n=20,
fillna=True)
data["dc_low_indicator"] = ta.donchian_channel_lband_indicator(close=data.close,
n=20,
fillna=True)
data["ichimoku_a"] = ta.ichimoku_a(high=data.high,
low=data.low,
n1=9,
n2=26,
fillna=True)
data.fillna(method="bfill")
# Normalizar los valores
for feature in data.columns:
norm_feature = preprocessing.normalize(data[feature].values.reshape(-1, 1), axis=0)
data[feature] = pd.DataFrame(data=norm_feature, index=data.index, columns=[feature])
self.assets = data.columns
trayectories.append(data.values)
closes.append(close)
return np.array(trayectories), np.array(closes)