def test_tsi2(self):
target = 'TSI'
result = tsi(**self._params)
pd.testing.assert_series_equal(self._df[target].tail(),
result.tail(),
check_names=False,
check_less_precise=True)
def get_ml_feature(self, symbol, prices=None, cutoff=None):
if prices:
price = prices.get(symbol, 1E10)
vix = prices['^VIX']
else:
price = self.closes[symbol][cutoff]
vix = self.closes['^VIX'][cutoff]
if cutoff:
close = self.closes[symbol][cutoff - DAYS_IN_A_YEAR:cutoff]
high = np.array(
self.hists[symbol].get('High')[cutoff - DAYS_IN_A_YEAR:cutoff])
low = np.array(
self.hists[symbol].get('Low')[cutoff - DAYS_IN_A_YEAR:cutoff])
else:
close = self.closes[symbol][-DAYS_IN_A_YEAR:]
high = np.array(self.hists[symbol].get('High')[-DAYS_IN_A_YEAR:])
low = np.array(self.hists[symbol].get('Low')[-DAYS_IN_A_YEAR:])
# Basic stats
day_range_change = price / np.max(close[-DATE_RANGE:]) - 1
today_change = price / close[-1] - 1
yesterday_change = close[-1] / close[-2] - 1
day_before_yesterday_change = close[-2] / close[-3] - 1
twenty_day_change = price / close[-20] - 1
year_high_change = price / np.max(close) - 1
year_low_change = price / np.min(close) - 1
all_changes = [
close[t + 1] / close[t] - 1 for t in range(len(close) - 1)
if close[t + 1] > 0 and close[t] > 0
]
# Technical indicators
close = np.append(close, price)
high = np.append(high, price)
low = np.append(low, price)
pd_close = pd.Series(close)
pd_high = pd.Series(high)
pd_low = pd.Series(low)
rsi = momentum.rsi(pd_close).values[-1]
macd_rate = trend.macd_diff(pd_close).values[-1] / price
wr = momentum.wr(pd_high, pd_low, pd_close).values[-1]
tsi = momentum.tsi(pd_close).values[-1]
feature = {
'Today_Change': today_change,
'Yesterday_Change': yesterday_change,
'Day_Before_Yesterday_Change': day_before_yesterday_change,
'Twenty_Day_Change': twenty_day_change,
'Day_Range_Change': day_range_change,
'Year_High_Change': year_high_change,
'Year_Low_Change': year_low_change,
'Change_Average': np.mean(all_changes),
'Change_Variance': np.var(all_changes),
'RSI': rsi,
'MACD_Rate': macd_rate,
'WR': wr,
'TSI': tsi,
'VIX': vix
}
return feature
def Feature_Extraction_Momentum(df):
df = SOK(df)
df = SOD(df)
df = SOJ(df)
df = MACD(df)
df = MACD_Signal(df)
df = MACD_Hist(df)
df['RSI'] = tas.RSI(df['close'].values)
df['ROC'] = tas.ROC(df['close'].values)
df['WillR'] = tas.WILLR(df['high'].values, df['low'].values,
df['close'].values)
df['CCI'] = tas.CCI(df['high'].values, df['low'].values,
df['close'].values)
df['TSI'] = tsi(df['close'], window_slow=25, window_fast=13)
df['ADX'] = tas.ADX(df['high'].values, df['low'].values,
df['close'].values)
df['MFI'] = tas.MFI(df['high'], df['low'], df['close'], df['volume'])
df['MOM'] = tas.MOM(df['close'])
df['TRIX'] = tas.TRIX(df['close'])
df['ULTOSC'] = tas.ULTOSC(df['high'], df['low'], df['close'])
df = df.fillna(0)
return df
def engineer_data_over_single_interval(df: pd.DataFrame,
indicators: list,
ticker: str = "",
rsi_n: int = 14,
cmo_n: int = 7,
macd_fast: int = 12,
macd_slow: int = 26,
macd_sign: int = 9,
roc_n: int = 12,
cci_n: int = 20,
dpo_n: int = 20,
cmf_n: int = 20,
adx_n: int = 14,
mass_index_low: int = 9,
mass_index_high: int = 25,
trix_n: int = 15,
stochastic_oscillator_n: int = 14,
stochastic_oscillator_sma_n: int = 3,
ultimate_oscillator_short_n: int = 7,
ultimate_oscillator_medium_n: int = 14,
ultimate_oscillator_long_n: int = 28,
ao_short_n: int = 5,
ao_long_n: int = 34,
kama_n: int = 10,
tsi_high_n: int = 25,
tsi_low_n: int = 13,
eom_n: int = 14,
force_index_n: int = 13,
ichimoku_low_n: int = 9,
ichimoku_medium_n: int = 26):
from ta.momentum import rsi, wr, roc, ao, stoch, uo, kama, tsi
from ta.trend import macd, macd_signal, cci, dpo, adx, mass_index, trix, ichimoku_a
from ta.volume import chaikin_money_flow, acc_dist_index, ease_of_movement, force_index
# Momentum Indicators
if Indicators.RELATIVE_STOCK_INDEX in indicators:
Logger.console_log(message="Calculating " +
Indicators.RELATIVE_STOCK_INDEX.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.RELATIVE_STOCK_INDEX.value] = rsi(close=df['close'],
n=rsi_n)
if Indicators.WILLIAMS_PERCENT_RANGE in indicators:
Logger.console_log(message="Calculating " +
Indicators.WILLIAMS_PERCENT_RANGE.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.WILLIAMS_PERCENT_RANGE.value] = wr(
df['high'], df['low'], df['close'])
if Indicators.CHANDE_MOMENTUM_OSCILLATOR in indicators:
Logger.console_log(message="Calculating " +
Indicators.CHANDE_MOMENTUM_OSCILLATOR.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.CHANDE_MOMENTUM_OSCILLATOR.
value] = chande_momentum_oscillator(close_data=df['close'],
period=cmo_n)
if Indicators.RATE_OF_CHANGE in indicators:
Logger.console_log(message="Calculating " +
Indicators.RATE_OF_CHANGE.value + " for stock " +
ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.RATE_OF_CHANGE.value] = roc(close=df['close'], n=roc_n)
if Indicators.STOCHASTIC_OSCILLATOR in indicators:
Logger.console_log(message="Calculating " +
Indicators.STOCHASTIC_OSCILLATOR.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.STOCHASTIC_OSCILLATOR.value] = stoch(
high=df['high'],
low=df['low'],
close=df['close'],
n=stochastic_oscillator_n,
d_n=stochastic_oscillator_sma_n)
if Indicators.ULTIMATE_OSCILLATOR in indicators:
Logger.console_log(message="Calculating " +
Indicators.ULTIMATE_OSCILLATOR.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.ULTIMATE_OSCILLATOR.value] = uo(
high=df['high'],
low=df['low'],
close=df['close'],
s=ultimate_oscillator_short_n,
m=ultimate_oscillator_medium_n,
len=ultimate_oscillator_long_n)
if Indicators.AWESOME_OSCILLATOR in indicators:
Logger.console_log(message="Calculating " +
Indicators.AWESOME_OSCILLATOR.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.AWESOME_OSCILLATOR.value] = ao(high=df['high'],
low=df['low'],
s=ao_short_n,
len=ao_long_n)
if Indicators.KAUFMAN_ADAPTIVE_MOVING_AVERAGE in indicators:
Logger.console_log(message="Calculating " +
Indicators.KAUFMAN_ADAPTIVE_MOVING_AVERAGE.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.KAUFMAN_ADAPTIVE_MOVING_AVERAGE.value] = kama(
close=df['close'], n=kama_n)
if Indicators.TRUE_STRENGTH_INDEX in indicators:
Logger.console_log(message="Calculating " +
Indicators.TRUE_STRENGTH_INDEX.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.TRUE_STRENGTH_INDEX.value] = tsi(close=df['close'],
r=tsi_high_n,
s=tsi_low_n)
# Trend Indicator
if Indicators.MOVING_AVERAGE_CONVERGENCE_DIVERGENCE in indicators:
Logger.console_log(
message="Calculating " +
Indicators.MOVING_AVERAGE_CONVERGENCE_DIVERGENCE.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.MOVING_AVERAGE_CONVERGENCE_DIVERGENCE.value] = macd(close=df['close'],
n_slow=macd_slow,
n_fast=macd_fast) - \
macd_signal(close=df['close'],
n_slow=macd_slow,
n_fast=macd_fast,
n_sign=macd_sign)
if Indicators.COMMODITY_CHANNEL_INDEX in indicators:
Logger.console_log(message="Calculating " +
Indicators.COMMODITY_CHANNEL_INDEX.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.COMMODITY_CHANNEL_INDEX.value] = cci(high=df['high'],
low=df['low'],
close=df['close'],
n=cci_n)
if Indicators.DETRENDED_PRICE_OSCILLATOR in indicators:
Logger.console_log(message="Calculating " +
Indicators.DETRENDED_PRICE_OSCILLATOR.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.DETRENDED_PRICE_OSCILLATOR.value] = dpo(
close=df['close'], n=dpo_n)
if Indicators.AVERAGE_DIRECTIONAL_INDEX in indicators:
Logger.console_log(message="Calculating " +
Indicators.AVERAGE_DIRECTIONAL_INDEX.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.AVERAGE_DIRECTIONAL_INDEX.value] = adx(high=df['high'],
low=df['low'],
close=df['close'],
n=adx_n)
if Indicators.MASS_INDEX in indicators:
Logger.console_log(message="Calculating " +
Indicators.MASS_INDEX.value + " for stock " +
ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.MASS_INDEX.value] = mass_index(high=df['high'],
low=df['low'],
n=mass_index_low,
n2=mass_index_high)
if Indicators.TRIPLE_EXPONENTIALLY_SMOOTHED_MOVING_AVERAGE in indicators:
Logger.console_log(
message="Calculating " +
Indicators.TRIPLE_EXPONENTIALLY_SMOOTHED_MOVING_AVERAGE.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.TRIPLE_EXPONENTIALLY_SMOOTHED_MOVING_AVERAGE.
value] = trix(close=df['close'], n=trix_n)
if Indicators.ICHIMOKU_A in indicators:
Logger.console_log(message="Calculating " +
Indicators.ICHIMOKU_A.value + " for stock " +
ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.ICHIMOKU_A.value] = ichimoku_a(high=df['high'],
low=df['low'],
n1=ichimoku_low_n,
n2=ichimoku_medium_n)
# Volume Indicator
if Indicators.CHAIKIN_MONEY_FLOW in indicators:
Logger.console_log(message="Calculating " +
Indicators.CHAIKIN_MONEY_FLOW.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.CHAIKIN_MONEY_FLOW.value] = chaikin_money_flow(
high=df['high'],
low=df['low'],
close=df['close'],
volume=df['volume'],
n=cmf_n)
if Indicators.ACCUMULATION_DISTRIBUTION_INDEX in indicators:
Logger.console_log(message="Calculating " +
Indicators.ACCUMULATION_DISTRIBUTION_INDEX.value +
" for stock " + ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.ACCUMULATION_DISTRIBUTION_INDEX.value] = acc_dist_index(
high=df['high'],
low=df['low'],
close=df['close'],
volume=df['volume'])
if Indicators.EASE_OF_MOVEMENT in indicators:
Logger.console_log(message="Calculating " +
Indicators.EASE_OF_MOVEMENT.value + " for stock " +
ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.EASE_OF_MOVEMENT.value] = ease_of_movement(
high=df['high'], low=df['low'], volume=df['volume'], n=eom_n)
if Indicators.FORCE_INDEX in indicators:
Logger.console_log(message="Calculating " +
Indicators.FORCE_INDEX.value + " for stock " +
ticker,
status=Logger.LogStatus.EMPHASIS)
df[Indicators.FORCE_INDEX.value] = force_index(close=df['close'],
volume=df['volume'],
n=force_index_n)
def get_ml_feature(self, symbol, prices=None, cutoff=None):
feature = {}
if prices:
price = prices.get(symbol, 1E10)
vix = prices['^VIX']
else:
price = self.closes[symbol][cutoff]
vix = self.closes['^VIX'][cutoff]
if cutoff:
close = self.closes[symbol][cutoff - DAYS_IN_A_YEAR:cutoff]
volume = self.volumes[symbol][cutoff - DAYS_IN_A_YEAR:cutoff]
else:
close = self.closes[symbol][-DAYS_IN_A_YEAR:]
volume = self.volumes[symbol][-DAYS_IN_A_YEAR:]
close = np.append(close, price)
# Log returns
feature['Day_1_Return'] = np.log(close[-1] / close[-2])
feature['Day_2_Return'] = np.log(close[-2] / close[-3])
feature['Day_3_Return'] = np.log(close[-3] / close[-4])
feature['Weekly_Return'] = np.log(price / close[-DAYS_IN_A_WEEK])
feature['Monthly_Return'] = np.log(price / close[-DAYS_IN_A_MONTH])
feature['Quarterly_Return'] = np.log(price / close[-DAYS_IN_A_QUARTER])
feature['From_Weekly_High'] = np.log(price /
np.max(close[-DAYS_IN_A_WEEK:]))
feature['From_Weekly_Low'] = np.log(price /
np.min(close[-DAYS_IN_A_WEEK:]))
# Technical indicators
pd_close = pd.Series(close)
feature['RSI'] = momentum.rsi(pd_close).values[-1]
feature['MACD_Rate'] = trend.macd_diff(pd_close).values[-1] / price
feature['TSI'] = momentum.tsi(pd_close).values[-1]
# Markets
feature['VIX'] = vix
# Other numerical factors
# Fit five data points to a second order polynomial
feature['Acceleration'] = (2 * close[-5] - 1 * close[-4] -
2 * close[-3] - 1 * close[-2] +
2 * close[-1]) / 14
feature['Momentum'] = (-2 * close[-5] - 1 * close[-4] + 1 * close[-2] +
2 * close[-1]) / 10
quarterly_returns = [
np.log(close[i] / close[i - 1])
for i in range(-DAYS_IN_A_QUARTER, -1)
]
monthly_returns = quarterly_returns[-DAYS_IN_A_MONTH:]
weekly_returns = quarterly_returns[-DAYS_IN_A_WEEK:]
feature['Monthly_Skewness'] = stats.skew(monthly_returns)
feature['Monthly_Volatility'] = np.std(monthly_returns)
feature['Weekly_Skewness'] = stats.skew(weekly_returns)
feature['Weekly_Volatility'] = np.std(weekly_returns)
feature['Z_Score'] = (
feature['Day_1_Return'] -
np.mean(quarterly_returns)) / np.std(quarterly_returns)
feature['Monthly_Avg_Dollar_Volume'] = np.average(
np.multiply(close[-DAYS_IN_A_MONTH - 1:-1],
volume[-DAYS_IN_A_MONTH:])) / 1E6
return feature