def preProcessCalculation(df, limit):
df = df.reset_index()
adx = np.array(
average_directional_index(df['close'], df['high'], df['low'], limit))
mfi = np.array(
money_flow_index(df['close'], df['high'], df['low'], df['volume'],
limit - 1))
atr = np.array(average_true_range(df['close'], limit))
obv = np.array(on_balance_volume(df['close'], df['volume']))
rocr = np.array(rate_of_change(df['close'], limit))
mom = np.array(momentum(df['close'], limit))
return adx, mfi, atr, obv, rocr, mom
def atr(dataframe, period, field='close') -> ndarray:
from pyti.average_true_range import average_true_range
return average_true_range(dataframe[field], period)
def test_average_true_range_period_8(self):
period = 8
atr = average_true_range.average_true_range(self.close_data, period)
np.testing.assert_allclose(atr, self.atr_period_8_expected)
def test_average_true_range_invalid_period(self):
period = 128
with self.assertRaises(Exception) as cm:
average_true_range.average_true_range(self.close_data, period)
expected = "Error: data_len < period"
self.assertEqual(str(cm.exception), expected)
def test_average_true_range_period_10(self):
period = 10
atr = average_true_range.average_true_range(self.close_data, period)
np.testing.assert_array_equal(atr, self.atr_period_10_expected)
def add_features(asset_data, features=None, save_path=None, verbose=True):
""" Adds features to the bar data. If no features are passed then all features
are added.
Args:
asset_data ((str, pd.DataFrame) or (str, pd.DataFrame)[]): a single
or list of tuples (asset name, bar data) containing the bar data
features (str[]): a list of features to include, all features if this
is None
save_path (str): Path to save the bar data with features. A
placeholder {ASSET} that will be substituted with the asset name
verbose (bool): True if progress printing to console is desired
Returns:
(str, pd.DataFrame): a single tuple of (asset name, dataframe) of the
bar data if a single asset was passed or an array of (asset name, dataframes)
if an array of assets was passed
"""
#helper function for creating logs
def replace_zero_with_min(series):
return series.replace(0, series.loc[series > 0].min())
import pandas as pd
import numpy as np
from pyti import bollinger_bands as bbands
from pyti import average_true_range as atr
#if a single dataframe is passed just put it in a single item list
if type(asset_data) == tuple:
asset_data = [asset_data]
elif type(asset_data) != list:
raise ValueError(
'asset_data must be a pandas.DataFrame or a list of pandas.Dataframe.'
)
feature_bars = []
for asset, data in asset_data:
if verbose: print("Calculating features for {0}".format(asset))
#Lower Bollinger Band, 20 periods, std = 2
if features == None or 'lower_bb' in features:
data['lower_bb'] = bbands.lower_bollinger_band(data["close"],
20,
std=2.0)
#Upper Bollinger Band, 20 periods, std = 2
if features == None or 'upper_bb' in features:
data['upper_bb'] = bbands.upper_bollinger_band(data["close"],
20,
std_mult=2.0)
#Average True Range
if features == None or 'atr' in features:
data["atr"] = atr.average_true_range(data["close"], 24) / \
atr.average_true_range(data["close"], 200)
#Volatility is the standard deviation over 12 periods of the difference
#between high and low of the bar
if features == None or 'volatility_12' in features:
data["volatility_12"] = (data["high"] -
data["low"]).rolling(12).std()
#Volatility is the standard deviation over 200 periods of the difference
#between high and low of the bar
if features == None or 'volatility_200' in features:
data["volatility_200"] = (data["high"] -
data["low"]).rolling(200).std()
#Volatility is relative change of the alst 12 bars over the last 200
if features == None or 'volatility' in features:
data["volatility"] = data["volatility_12"] / data["volatility_200"]
#Relative volume compared to the last 100 bars
if features == None or 'volume_change' in features:
data["volume_change"] = data["volume"] / data["volume"].rolling(
100).mean()
#Distance between the close price and the upper bollinger bad
if features == None or 'bb_dist_upper' in features:
data["bb_dist_upper"] = data["upper_bb"] - data["close"]
#Distance between the close price and the lower bollinger bad
if features == None or 'bb_dist_lower' in features:
data["bb_dist_lower"] = -(data["lower_bb"] - data["close"])
#Distance between the upper and lower bollinger bands
if features == None or 'bb_range' in features:
data["bb_range"] = (data["upper_bb"] -
data["lower_bb"]) / data["close"]
#The absolute value of the % return over the last 4 bars
if features == None or 'change_4bar' in features:
data["change_4bar"] = np.abs(
np.log(data["close"] / data["close"].shift(4)))
#The Augmented Dicker Fuller test which can show mean reverting or trending markets
#from statsmodels.tsa.stattools import adfuller
#if features == None or 'adf' in features:
# data["adf"] = data['Close'].rolling(200).apply(lambda x: adfuller(x)[0], raw=False)
#The log of the % return
if features == None or 'log_return' in features:
data["log_return"] = np.log(data["close"] / data["close"].shift(1))
#TODO add in the if feature statements
#add logs - do this by replacing all zeros with the minimum value after zeros are removed
if 1 == 0:
data["volume_log"] = np.log(replace_zero_with_min(
data["volume"])) / np.log(
replace_zero_with_min(data["volume"].rolling(200).mean()))
data["atr_log"] = np.log(replace_zero_with_min(data["atr"]))
data["volatility_log"] = np.log(
replace_zero_with_min(data["volatility"]))
data["change_4bar_log"] = np.log(
replace_zero_with_min(data["change_4bar"]))
#add moving averages
data["volume_log_ma_12"] = data["volume_log"].rolling(12).mean()
data["volume_log_ma_24"] = data["volume_log"].rolling(24).mean()
data["volume_log_ma_48"] = data["volume_log"].rolling(48).mean()
data["volatility_log_ma_12"] = data["volatility_log"].rolling(
12).mean()
data["volatility_log_ma_24"] = data["volatility_log"].rolling(
24).mean()
data["volatility_log_ma_48"] = data["volatility_log"].rolling(
48).mean()
data["change_4bar_log_ma_12"] = data["change_4bar_log"].rolling(
12).mean()
data["change_4bar_log_ma_24"] = data["change_4bar_log"].rolling(
24).mean()
data["change_4bar_log_ma_48"] = data["change_4bar_log"].rolling(
48).mean()
data["log_return_ma_12"] = data["log_return"].rolling(12).mean()
data["log_return_ma_24"] = data["log_return"].rolling(24).mean()
data["log_return_ma_48"] = data["log_return"].rolling(48).mean()
feature_bars.append((asset, data))
if save_path is not None:
if verbose: print("Saving {0}...".format(asset))
save_location = save_path.replace("{ASSET}", asset)
data.to_csv(save_location)
#if there was just one dataFrame just return a single dataframe, otherwise
#return the list of dataframes, one for each asset
if len(feature_bars) == 1:
return feature_bars[0]
else:
return feature_bars
