def run_scenario(assets, sigma, runid, prices, results):
excess_returns = rr.calc_excess_returns(prices,
msv.default_parameters.risk_free_rate)
is_y_feasible = True
y = 0.
result_path = results + "/" + runid
filename = result_path + 'out'
assets = map(int, assets.split())
write_portfolio_header(filename, assets)
write_time_file_header(filename)
scenario = Scenario(assets, excess_returns)
# Search from y=0. until infeasible, possibly skipping the
# semivar_bound value entirely if y=0. is infeasible.
while is_y_feasible:
try:
t0 = time.time()
u_star = calc_dcc_portfolio(y, sigma, scenario.train, scenario.model)
except ZeroDivisionError as e:
print 'Divide by Zero Error: {0}'.format(e.message)
print 'Skipping y = {0}, sigma = {1}'.format(y, sigma)
is_y_feasible = False
except np.linalg.linalg.LinAlgError as e:
print 'Linear Algebra Error: {0}'.format(e.message)
print 'Skipping y = {0}, sigma = {1}'.format(y, sigmaf)
is_y_feasible = False
else:
t1 = time.time()
run_time = t1 - t0
print 'u_star = %r' % u_star
print 'Run time = %r' % run_time
append_portfolio(filename, assets, u_star, y, sigma)
append_time_file(filename, assets, y, sigma, run_time)
y = log_increment(y)
is_y_feasible = y < 100.
import numpy.linalg as la
def load_output(directory):
output = list()
files = os.listdir(directory)
for filename in files:
if re.search('out$', filename):
output.append(pd.read_csv(os.path.join(directory,filename)))
return output
results = './results/20160320'
p = load_output(results)
excess_returns = rr.calc_excess_returns('./data/raw/yahoo/jiang_guan_daily_close_29assets.csv',
1.0,
use_monthly=True)
metrics = list()
i = 0
for pi in p:
print i
metrics.append(dict())
metrics[i]['assets'] = map(int, pi.assets[0].split(' '))
metrics[i]['scenario'] = scen.Scenario(metrics[i]['assets'], excess_returns)
metrics[i]['mv'] = mv.portfolio(metrics[i]['scenario'].train, 1.0)
metrics[i]['dcc'] = pi[['u({0})'.format(j) for j in range(4)]].values
metrics[i]['dcc_train_mean'] = np.apply_along_axis(
lambda x: ps.mean(x, metrics[i]['scenario'].train), 1, metrics[i]['dcc'])
opt_idx = np.argmax(metrics[i]['dcc_train_mean'])
