def _update(self, weight: np.ndarray, grad: np.ndarray, cache: Dict[str, Any]) -> Dict[str, Any]:
g_avg_t_minus_1 = cache.get('g_avg', np.zeros_like(grad))
update_avg_t_minus_1 = cache.get('u_avg', np.zeros_like(grad))
g_avg = linear_combination(g_avg_t_minus_1, grad**2, self._delta)
update = grad * ((np.sqrt(update_avg_t_minus_1 + self._eps)) / np.sqrt(g_avg + self._eps))
update_avg = linear_combination(update_avg_t_minus_1, update**2, self._delta)
weight -= update
return {'g_avg': g_avg, 'u_avg': update_avg}
def predict(self, X, Y=None):
data = self._data_fetcher(X, last=True)
df_features = create_features(data)
df_features = df_features.round(2)
X_train, X_test, y_train, y_test = create_X_Y(df_features)
self.lr.fit(X_train, y_train)
y_pred = self.lr.predict(X_test)
y_pred = y_pred.round(2)
score = self.lr.score(X_test, y_test)
mean_absolute_error = metrics.mean_absolute_error(y_test, y_pred)
mean_squared_error = metrics.mean_squared_error(y_test, y_pred)
sqrt_mean_squared_error = np.sqrt(
metrics.mean_squared_error(y_test, y_pred))
predictions = self.lr.predict(X_train)
predictions = predictions.round(2)
prediction_for_y_train = predictions.flatten()[-1]
# shape = df_features.shape()
# #Classification
# for i in range(1,shape[1]+1):
# if df_features[i]:
#
# df_features['buysell']
return {
"prediction_for_y_train": prediction_for_y_train,
"r2_score": score,
"mean_absolute_error": mean_absolute_error,
"mean_squared_error": mean_squared_error,
"sqrt_mean_squared_error": sqrt_mean_squared_error,
}
def range_volatility(ticker, components):
# Parse and check: find components matching dates, ensure start and end are present, ensure dates are valid
today, dates = current_date(), []
for each in components:
if PATTERNS['valid_date'].match(each):
dates.append(each)
if len(dates)!=2:
return {"message": Response.vol_required_dates(ticker)}
for each in dates:
if each > today:
return {"message": Response.invalid_date(each)}
try:
date = datetime.datetime.strptime(each, '%Y-%m-%d')
except ValueError:
return {"message": Response.invalid_date(each)}
# Volatility Calculation
dates = sorted(dates)
start, end = dates[0], dates[1]
try:
quotes = data.DataReader(ticker, 'google')['Close'].loc[start:end]
if len(quotes) < 10:
return {"message": Response.vol_range_size(ticker)}
except Exception:
return {"message": Response.data_notfound(ticker)}
logreturns = np.log(quotes / quotes.shift(1))
vol = round(np.sqrt(252*logreturns.var()), 5)
return {"message" : Response.range_vol(ticker, start, end, vol)}
def _setPortfolioInfo(self):
port_annual_var: float = round(np.dot(self._weights.T, np.dot(self._dataSimpleCovarianceAnnual, self._weights)), 5)
port_annual_volatility: float = round(np.sqrt(port_annual_var), 5)
port_annual_simple_ret: float = round(np.sum(self._stats.SimpleReturnsNan.mean() * self._weights) * 252, 5)
print('Port Ann Ret', str(round(port_annual_var, 5)*100)+'%')
print('Port Ann Volatility/ Risk', str(round(port_annual_volatility, 5)*100)+'%')
print('Port Ann Variance', str(round(port_annual_simple_ret, 5)*100)+'%')
'''
def hist_vol(sym, days=10):
try:
quotes = DataReader(sym, 'yahoo')['Close'][-days:]
except Exception:
print "Problem getting historical volatility!"
raise SystemExit(code)
return None, None
logreturns = np.log(quotes / quotes.shift(1))
vol = np.sqrt(252*logreturns.var()) #252 trading days in year (annualized volatility)
return float(vol)
def trailing_volatility(ticker, days):
"""
Returns equally-weighted historical volatility for a stock during the last X trading days
(annualized std deviation of daily log returns) using market close prices.
Google API - data available from Jan 4, 2010 to present and is adjusted for stock splits.
"""
try:
ticker_input = "WIKI/" + ticker
quotes = quandl.get(ticker_input, order='desc')['Close'][:days]
except Exception:
return False
logreturns = np.log(quotes / quotes.shift(1))
return round(np.sqrt(252 * logreturns.var()), 5)
def trailing_volatility(ticker, components):
days=None
for each in components:
if PATTERNS['tvol'].match(each):
days = int(each)
break
if days==None:
return {"message": Response.trailing_days(ticker)}
try:
quotes = data.DataReader(ticker, 'google')['Close'][-days:]
except Exception:
return {"message": Response.data_notfound(ticker)}
logreturns = np.log(quotes / quotes.shift(1))
vol = round(np.sqrt(252*logreturns.var()), 5)
return {"message" : Response.trailing_vol(days, ticker, vol)}
def PlotMatrix(self) -> plt:
prf_returns = (self._stats.Returns.pct_change() + 1)[1:]
avg_return = (prf_returns-1).mean()
daily_pct_change = np.log(self._stats.Returns.pct_change() + 1)
vols = daily_pct_change.std() * np.sqrt(252)
plt.style.use('seaborn')
plt.rcParams['date.epoch'] = '0000-12-31'
fig, ax = plt.subplots(1, 1, figsize=(self._a_float, self._a_float/2.0))
#print('AX', type(ax)) AX <class 'matplotlib.axes._subplots.AxesSubplot'>
fig.suptitle(self._basics.Title)
ax.set(ylabel='Simple Return Std', xlabel='Simple Return Mean')
ax.scatter(vols, avg_return)
for i, txt in enumerate(list(vols.index)):
ax.annotate(txt, (vols[i], avg_return[i]))
plt.tight_layout()
#plt.show()
return plt
def _setMatrices(self, portfolio_data: DataFrame, log_ret: DataFrame,
cov_mat: DataFrame):
for i in range(self._threshold):
weight_arr: np.ndarray = np.random.uniform(
size=len(portfolio_data.columns))
weight_arr = weight_arr / np.sum(weight_arr)
# saving weights in the array
self._weight_matrix[i, :] = weight_arr
# Portfolio Returns
annual_weighted_log_ret: float = (
(np.sum(log_ret.mean() * weight_arr)) + 1)**252 - 1
# Saving Portfolio returns
self._annual_weighted_log_return_matrix[
i] = annual_weighted_log_ret
# Saving Portfolio Risk
portfolio_sd: float = np.sqrt(
np.dot(weight_arr.T, np.dot(cov_mat, weight_arr)))
self._risk_matrix[i] = portfolio_sd
# Portfolio Sharpe Ratio
# Assuming 0% Risk Free Rate
sr: float = annual_weighted_log_ret / portfolio_sd
self._sharpe_ratio_matrix[i] = sr
def _getAnnualReturnStd(self, a_series: Series = Series()) -> float:
return self.__roundFloat(np.std(a_series) * np.sqrt(252))
from pandas import np
import pandas_datareader.data as web
def historical_volatility(sym, days): # stock symbol, number of days
"Return the annualized stddev of daily log returns of picked stock"
try:
# past number of 'days' close price data, normally between (30, 60)
quotes = web.DataReader(sym, 'yahoo')['Close'][-days:]
except Exception, e:
print "Error getting data for symbol '{}'.\n".format(sym), e
return None, None
logreturns = np.log(quotes / quotes.shift(1))
# return square root * trading days * logreturns variance
# NYSE = 252 trading days; Shanghai Stock Exchange = 242; Tokyo Stock Exchange = 246 days?
return np.sqrt(252*logreturns.var())
if __name__ == "__main__":
print historical_volatility('FB', 30) # facebook
def wgn(X, snr):
P_signal = np.sum(abs(X)**2) / (len(X))
P_noise = P_signal / 10**(snr / 10.0)
#return np.random.randn(X.shape[1])*np.sqrt(P_noise)
return np.random.randn(len(X)) * np.sqrt(P_noise)
def __init__(self, y_stocks: list):
self._a_float = 3 * math.log(y_stocks[0].TimeSpan.MonthCount)
self._a_suffix = y_stocks[0].Column
self._a_ts = y_stocks[0].TimeSpan
self._a_length = len(y_stocks)
iso_weight: float = round(1.0 / len(y_stocks), 3)
self._stocks = y_stocks
self._weights = np.array(len(y_stocks) * [iso_weight], dtype=float)
self._basics = PortfolioBasics(y_stocks, self._a_float, self._legend_place)
self._stats = PortfolioStats(self._weights, self._basics)
self._final = PortfolioFinal(y_stocks, self._a_float, self._legend_place)
print('Volatility\t\t\t\t\t', self._final.Volatility)
print('Annual Expected Return\t\t', self._final.AnnualExpectedReturn)
print('Risk Free Rate\t\t\t\t', self._final.RiskFreeRate)
print('Free 0.005 Sharpe Ratio\t\t', self._final.Free005SharpeRatio)
print('Kurtosis\n', self._final.KurtosisSeries)
print('Skewness\n', self._final.SkewnessSeries)
print('Frequency\n', self._final.Frequency)
self._final.Plot().show()
exit(1234)
self._dataSimpleCorrelation = self._stats.SimpleReturnsNan.corr()
self._dataSimpleCovariance = self._stats.SimpleReturnsNan.cov()
self._dataSimpleCovarianceAnnual = self._dataSimpleCovariance * 252
self._dataSimpleSummary = self._stats.SimpleReturnsNanSummary
self._dataWeightedReturns = self._stats.SimpleWeightedReturns
# axis =1 tells pandas we want to add the rows
self._portfolio_weighted_returns = round(self._dataWeightedReturns.sum(axis=1), 5)
print('7', self._portfolio_weighted_returns.head())
print('7', self._stats.SimpleWeightedReturnsSum.head())
#self._dataWeightedReturns['PORTFOLIOWeighted'] = portfolio_weighted_returns
portfolio_weighted_returns_mean = round(self._portfolio_weighted_returns.mean(), 5)
print('port_ret mean', portfolio_weighted_returns_mean)
print(round(self._stats.SimpleWeightedReturnsSum.mean(), 5))
portfolio_weighted_returns_std = round(self._portfolio_weighted_returns.std(), 5)
print('port_ret std', portfolio_weighted_returns_std)
self._portfolio_weighted_returns_cum: Series = round((self._portfolio_weighted_returns + 1).cumprod(), 5)
#self._dataWeightedReturns['PORTFOLIOCumulative'] = self._portfolio_weighted_returns_cum
print('$', self._dataWeightedReturns.head())
self._portfolio_weighted_returns_geom = round(np.prod(self._portfolio_weighted_returns + 1) ** (252 / self._portfolio_weighted_returns.shape[0]) - 1, 5)
print('geometric_port_return', self._portfolio_weighted_returns_geom)
self._portfolio_weighted_annual_std = round(np.std(self._portfolio_weighted_returns) * np.sqrt(252), 5)
print('port_ret annual', self._portfolio_weighted_annual_std)
self._portfolio_weighted_sharpe_ratio = round(self._portfolio_weighted_returns_geom / self._portfolio_weighted_annual_std, 5)
print('port_sharpe_ratio', self._portfolio_weighted_sharpe_ratio)
print('%', self._stats.Returns.head())
self._data_returns_avg = self._getDataReturnsAverage(self._stats.Returns)
print('^', self._data_returns_avg.head())
daily_log_pct_changes: DataFrame = np.log(self._stats.Returns.pct_change() + 1) #avant portfolio
daily_log_pct_changes.columns = daily_log_pct_changes.columns + 'LogReturn'
print('&', daily_log_pct_changes.head())
daily_log_volatilities: DataFrame = (daily_log_pct_changes.std() * np.sqrt(252)).to_frame()
daily_log_volatilities.columns = ['Volatility']
print('*', daily_log_volatilities)
port_daily_simple_ret: float = round(np.sum(self._stats.SimpleReturnsNan.mean()*self._weights), 5)
port_weekly_simple_ret: float = round(4.856 * port_daily_simple_ret, 5)
port_monthly_simple_ret: float = round(21 * port_daily_simple_ret, 5)
port_quarterly_simple_ret: float = round(63 * port_daily_simple_ret, 5)
port_yearly_simple_ret: float = round(252 * port_daily_simple_ret, 5)
print('port_daily_simple_ret', str(100*port_daily_simple_ret) + '%')
print('port_weekly_simple_ret', str(100*port_weekly_simple_ret) + '%')
print('port_monthly_simple_ret', str(100*port_monthly_simple_ret) + '%')
print('port_quarterly_simple_ret', str(100*port_quarterly_simple_ret) + '%')
print('port_yearly_simple_ret', str(100*port_yearly_simple_ret) + '%')
self._setPortfolioInfo()
self._optimizer = PortfolioOptimizer(self._legend_place, self._a_float, self._stats, self._basics.Data)
self._stock_market_index = SnP500Index('yahoo', "^GSPC", self._a_ts)
self._linear_reg = PortfolioLinearReg(self._stock_market_index, self._stats.Returns)
print(f'The portfolio beta is {self._linear_reg.Beta}, for each 1% of index portfolio will move {self._linear_reg.Beta}%')
print('The portfolio alpha is ', self._linear_reg.Alpha)
print('_', self._basics.DataLogReturns.head())
cov_mat_annual = self._basics.DataLogReturns.cov() * 252
print('-', cov_mat_annual)
def calculate(param, param1, param2, param3):
xDis = abs(param - param2)
yDis = abs(param1 - param3)
from pandas import np
distance = np.sqrt((xDis ** 2) + (yDis ** 2))
return int(distance)
def order_path(gray_img, locs, xy_to_pixel):
# starting location
# TODO: pass this in dynamically
start_loc = locs[0]
# construct 2d cost map where each entry is cost to move from neighboring pixel
h, w = np.shape(gray_img)
# iterate over the entire image and build graph for dijkstra calculations
print('Converting map into graph.')
graph = Graph()
# NOTE: this isn't using the values in the map yaml (0.196) for free_thresh but instead assumes
# 0 = occupied, 205 = unknown, 254 = free
free_thresh = 220
diag_cost = np.sqrt(2)
lin_cost = 1
for py in range(1, h - 1):
for px in range(1, w - 1):
if gray_img[py][px] < free_thresh:
# obstacle -- skip it
continue
i = py * w + px
for py2 in range(py - 1, py + 2):
for px2 in range(px - 1, px + 2):
if gray_img[py2][px2] < free_thresh:
# obstacle -- skip it
continue
if px == px2 and py == py2:
# same as "center" pixel -- skip it
continue
i2 = py2 * w + px2
if px == px2 or py == py2:
# straight up or down pixel
graph.add_edge(i, i2, lin_cost)
else:
# diag pixel
graph.add_edge(i, i2, diag_cost)
# perform dijkstra cost calculation for each disinfection location pair
# to build cost array for traveling salesman problem
print('Building cost matrix for TSP problem.')
num_locs = len(locs)
costs = [[0] * num_locs for i in range(num_locs)]
for loc_i in range(num_locs):
costs[loc_i][loc_i] = 0
(px, py) = xy_to_pixel(locs[loc_i][0], locs[loc_i][1])
src_i = py * w + px
for loc_j in range(loc_i + 1, num_locs):
(px, py) = xy_to_pixel(locs[loc_j][0], locs[loc_j][1])
dst_i = py * w + px
path_info = find_path(graph, src_i, dst_i)
cost = int(round(path_info.total_cost))
costs[loc_i][loc_j] = cost
costs[loc_j][loc_i] = cost
# find disinfection location closest to and furthest from starting location (i.e., where robot currently is)
# this will serve as the start and end locations for the tsp problem
print('Finding starting and ending disinfection locations.')
(px, py) = xy_to_pixel(start_loc[0], start_loc[1])
src_i = py * w + px
start_i = -1
end_i = -1
min_cost = sys.maxsize
max_cost = 0
for loc_i in range(num_locs):
(px, py) = xy_to_pixel(locs[loc_i][0], locs[loc_i][1])
dst_i = py * w + px
path_info = find_path(graph, src_i, dst_i)
cost = int(round(path_info.total_cost))
if cost < min_cost:
min_cost = cost
start_i = loc_i
if cost > max_cost:
max_cost = cost
end_i = loc_i
# feed cost array into tsp algorithm
print('Solving TSP problem.')
data = create_data_model(costs, start_i, end_i)
tour = solve_tsp(data)
return tour
from pandas import np
from pandas_datareader.data import DataReader
def ewma_volatility(sym, days):
try:
quotes = DataReader(sym, 'google')['Close'][-days:]
except Exception, e:
print "Error getting data for symbol '{}'.\n".format(sym), e
return None, None
logreturns = np.log(quotes / quotes.shift(1))
l = 0.94 #lambda
temp = 1 - l
weight = []
for logreturn in logreturns:
logreturn = logreturn * logreturn * temp
weight.append(logreturn)
temp = temp * l
ewma = pandas.Series(weight, logreturns.index)
return np.sqrt(252 * ewma.sum())
if __name__ == "__main__":
print ewma_volatility('GOOG', int(sys.argv[1])) * 100
# @Last Modified by: boyac
# @Last Modified time: 2016-05-02 19:09:29
from pandas import np
import pandas_datareader.data as web
def gk_vol(sym, days):
""""
Return the annualized stddev of daily log returns of picked stock
Historical Open-High-Low-Close Volatility: Garman Klass
sigma**2 = ((h-l)**2)/2 - (2ln(2) - 1)(c-o)**2
ref: http://www.wilmottwiki.com/wiki/index.php?title=Volatility
"""
try:
o = web.DataReader(sym, 'yahoo')['Open'][-days:]
h = web.DataReader(sym, 'yahoo')['High'][-days:]
l = web.DataReader(sym, 'yahoo')['Low'][-days:]
c = web.DataReader(sym, 'yahoo')['Close'][-days:]
except Exception, e:
print "Error getting data for symbol '{}'.\n".format(sym), e
return None, None
sigma = np.sqrt(252 * sum((np.log(h / l))**2 / 2 - (2 * np.log(2) - 1) *
(np.log(c / o)**2)) / days)
return sigma
if __name__ == "__main__":
print gk_vol('FB', 30) # facebook
data = data_fetcher(ticker, last=True)
data = data.drop('ticker', axis=1)
data = data.round(2)
X = data.drop('close', axis=1)
Y = data[['close']]
X_train, X_test, y_train, y_test = train_test_split(X,
Y,
test_size=test_size)
regressor = LinearRegression()
regressor.fit(X_train, y_train)
y_pred = regressor.predict(X_test)
y_pred = y_pred.round(2)
mean_absolute_error = metrics.mean_absolute_error(y_test, y_pred)
mean_squared_error = metrics.mean_squared_error(y_test, y_pred)
sqrt_mean_squared_error = np.sqrt(
metrics.mean_squared_error(y_test, y_pred))
y_pred_from_train = regressor.predict(X_train)
y_pred_from_train = y_pred_from_train.round(2)
# last_close_price = get_last_close_price(ticker)
print(f'Prediction for ticker {ticker}: {y_pred_from_train}'
f'\n\nScore given with test_size {test_size} is:'
f'\n\tmean_absolute_error": {mean_absolute_error}'
f'\n\tmean_squared_error: {mean_squared_error}'
f'\n\tsqrt_mean_squared_error": {sqrt_mean_squared_error}'
# f'\n\nLast close price was: {last_close_price.get("close")}'
)
def volatility(df):
quotes = df['close']
logreturns = np.log(quotes / quotes.shift(1))
vol = np.sqrt(252 * logreturns.var())
return vol
