def mean_reversion_signal(data, hl, window):
if 'mid_price' not in data.columns:
add_mid_price(data)
ema = pd.ewma(data['mid_price'], halflife=hl)
std = pd.rolling_std(data['mid_price'], window=window, min_periods=0)
data['mean_reversion_signal_{}_{}'.format(hl, window)] = (data['mid_price'] - ema) / std
return ['mean_reversion_signal_{}_{}'.format(hl, window)]
def dema(data, window, do_raise=True):
if 'mid_price' not in data.columns:
add_mid_price(data)
colname = 'dema_{}'.format(window)
ma = pd.ewma(data['mid_price'], halflife=window)
data[colname] = np.concatenate([[0], np.diff(ma)])
return [colname]
def dma(data, window, do_raise=True):
if 'mid_price' not in data.columns:
add_mid_price(data)
colname = 'dma_{}'.format(window)
ma = pd.rolling_mean(data['mid_price'], window=window, min_periods=0)
data[colname] = np.concatenate([[0], np.diff(ma)])
return [colname]
def future_profit_std(data, window, do_raise=True):
if 'mid_price' not in data.columns:
add_mid_price(data)
dprice = np.concatenate([[0], np.diff(data['mid_price'].values)])
data['profit_std_{}w+'.format(window)] = np.concatenate([
pd.rolling_std(dprice[::-1], window=window, min_periods=0)[:0:-1], [0]
])
return ['profit_std_{}w+'.format(window)]
def dema_diffs(data, windows, do_raise=True):
names = []
if 'mid_price' not in data.columns:
add_mid_price(data)
for window in windows:
ma = pd.ewma(data['mid_price'], halflife=window)
data['dema_{}'.format(window)] = np.concatenate([[0], np.diff(ma)])
for w1 in windows:
for w2 in windows:
if w1 > w2:
feat_name = 'dema_diff_{}-{}'.format(w1, w2)
data[feat_name] = data['dema_{}'.format(w1)] - data['dema_{}'.format(w2)]
names.append(feat_name)
return names
def dma_diffs(data, windows, do_raise=True):
names = []
if 'mid_price' not in data.columns:
add_mid_price(data)
for window in windows:
ma = pd.rolling_mean(data['mid_price'], window=window, min_periods=0)
data['dma_{}'.format(window)] = np.concatenate([[0], np.diff(ma)])
for w1 in windows:
for w2 in windows:
if w1 > w2:
feat_name = 'dma_diff_{}-{}'.format(w1, w2)
data[feat_name] = data['dma_{}'.format(w1)] - data['dma_{}'.format(w2)]
names.append(feat_name)
return names
def ema_diff(data, hl):
if 'mid_price' not in data.columns:
add_mid_price(data)
ema = pd.ewma(data['mid_price'], halflife=hl)
data['ema_diff_{}'.format(hl)] = data['mid_price'] - ema
return ['ema_diff_{}'.format(hl)]
