def __init__(self, ticker, over=7):
super(Stock, self).__init__()
self.ticker = ticker
self.over = over
self.rescale = MaxRescale(
['Adjusted Close', 'Volume', 'Open', 'Low', 'High', 'SP Close'])
self.bollinger10 = BollingerBand(window=10)
self.bollinger5 = BollingerBand(window=5)
class Stock(object):
"""docstring for Stock"""
def __init__(self, ticker, over=7):
super(Stock, self).__init__()
self.ticker = ticker
self.over = over
self.rescale = MaxRescale(
['Adjusted Close', 'Volume', 'Open', 'Low', 'High', 'SP Close'])
self.bollinger10 = BollingerBand(window=10)
self.bollinger5 = BollingerBand(window=5)
def get_data(self, start_date, end_date, fit=False, cache=True):
d_file_name = "d{}_{}_{}.csv".format(start_date, end_date, self.ticker)
l_file_name = "l{}_{}_{}.csv".format(start_date, end_date, self.ticker)
if cache and os.path.exists(d_file_name) and os.path.exists(l_file_name):
data = pd.read_csv(d_file_name, index_col=0, parse_dates='Date')
label = pd.read_csv(
l_file_name,
index_col=0,
parse_dates=True,
squeeze=True,
header=None)
else:
data = QuandlAPI.get_data(self.ticker, start_date, end_date, 160, 15)
label = Return(self.over).transform(data)
#######################
# Augment with index #
#######################
sp500 = QuandlAPI.get_data(
"INDEX_GSPC", start_date, end_date, 75, 15)
data = data.join(sp500['Adj Close'], how='left')
data.rename(columns={'Adj Close': 'SP Close'}, inplace=True)
# Deal with outliers
desc = data.describe()
iqr = desc.ix['75%'] - desc.ix['25%']
data = data[data <= desc.ix['50%'] + 1.5*iqr]
data = data[data >= desc.ix['50%'] - 1.5*iqr]
# Cache the raw data
if cache:
data.to_csv(d_file_name, index=True)
label.to_csv(l_file_name, index=True)
##########
# Adjust #
##########
# TODO There may be a problem when subsequent call are made
# The adjusted is always according to the last date
# So two subsequent call with two different period will be adjusted
# differently. Not sure if it matters or not
ratio = data['Adjusted Close'] / data['Close']
data['High'] = data['High']*ratio
data['Low'] = data['Low']*ratio
data['Open'] = data['Open']*ratio
##################
# Rescale by max #
##################
if fit:
self.rescale.fit(data, start_date, end_date)
data = self.rescale.transform(data)
######################
# Average True Range #
######################
# pdb.set_trace()
data['ATR'] = ATR(data, 14)
##################
# Bollinger band #
##################
data = self.bollinger10(data)
# data = self.bollinger5(data)
#################
# Rolling stats #
#################
data = high(22, data) # Chandelier
data = low(22, data) # Chandelier
# Ichimoku Clouds
# http://stockcharts.com/school/doku.php?id=chart_school:technical_indicators:ichimoku_cloud
data['tenkansen9'] = sen(data, 9)
data['kijunsen26'] = sen(data, 26)
data['senkouA'] = (data['tenkansen9'] + data['kijunsen26']) / 2. # Not sure if redondant. Affine transformation of two other features
data['senkouB52'] = sen(data, 52)
data['chikou26'] = data['Adjusted Close'].shift(26)
# Moving Average
# http://stockcharts.com/school/doku.php?id=chart_school:technical_indicators:moving_averages
data['ema10'] = ema(data, 10)
data['ema10-100'] = ema(data, 10) / ema(data, 100)
data['ema25-100'] = ema(data, 25) / ema(data, 100)
data['ema50-100'] = ema(data, 50) / ema(data, 100)
data['kama'] = kama(data)
data['ret_over2'] = return_over(data, 2)
# RSI
# http://stockcharts.com/school/doku.php?id=chart_school:technical_indicators:relative_strength_index_rsi
data['rsi'] = rsi(data)
# Price Volume Oscillator
# http://stockcharts.com/school/doku.php?id=chart_school:technical_indicators:percentage_volume_oscillator_pvo
ema_vol10 = ema(data, 10, 'Volume')
ema_vol26 = ema(data, 26, 'Volume')
data['pvo'] = (ema_vol10-ema_vol26)/ema_vol10
# This is to make sure that both index are aligned.
# If the start_date and end_date don't play well with the features
# I'm making sure that it does not screw up the rest
data.dropna(inplace=True)
label.dropna(inplace=True)
data, label = data[start_date:end_date], label[start_date:end_date]
data_index, label_index = data.index, label.index
max_start = max(data_index[0], label_index[0])
min_end = min(data_index[-1], label_index[-1])
data, label = data[max_start: min_end], label[max_start: min_end]
del data['Close']
# del data['Volume']
# pdb.set_trace()
return data, label