def parallel_optimization(num, portfolios, portfolio, all_stocks, iterations,
type_ga):
stocks = all_stocks[num]
days = len(stocks[0].day_history)
cov_matrix, mean_daily_returns, avr_gmean, gmeans = cov_and_mean(stocks)
problemGenerator = NSGAII.PortfolioGenerator(portfolio)
start = time.time()
if type_ga == GA_SIMPLE:
results_frame = simple.solve(stocks, iterations, mean_daily_returns,
cov_matrix, days)
elif type_ga == GA_NSGAII:
results_frame = NSGAII.solve(stocks,
iterations,
mean_daily_returns,
cov_matrix,
days,
generator=problemGenerator)
else:
results_frame = NSGAII.solve_nsgaiii(stocks,
iterations,
mean_daily_returns,
cov_matrix,
days,
generator=problemGenerator)
duration = time.time() - start
LOG.info('Duration solved: %s' % duration)
new_sharpe, new_id = process_result_of_ga(results_frame, stocks, avr_gmean)
LOG.info('Solve %d of %d' % (num + 1, len(portfolios)))
return new_sharpe, new_id
def parallel_solve(all_stocks, type_ga, curr, count):
LOG.info('Start parallel solve in %s' % str(threading.get_ident()))
range_stock = len(all_stocks)
number = list()
stocks = []
for position in range(15):
pos = random.randint(0, range_stock)
while pos in number:
pos = random.randint(0, range_stock)
stock = all_stocks[pos - 1]
while len(stock.day_history) < 66:
pos = random.randint(0, range_stock)
stock = all_stocks[pos - 1]
stocks.append(stock)
# download daily price data for each of the stocks in the portfolio
data = get_stock_price(stocks)
returns = data.pct_change()
# calculate mean daily return and covariance of daily returns
mean_daily_returns = returns.mean()
cov_matrix = returns.cov()
days = len(stocks[0].day_history)
iterations = 50000
start = time.time()
if type_ga == GA_SIMPLE:
results_frame = simple.solve(stocks, iterations, mean_daily_returns,
cov_matrix, days)
if type_ga == GA_NSGAII:
results_frame = NSGAII.solve(stocks, iterations, mean_daily_returns,
cov_matrix, days)
if type_ga == GA_NSGAIII:
results_frame = NSGAII.solve_nsgaiii(stocks, iterations,
mean_daily_returns, cov_matrix,
days)
duration = time.time() - start
LOG.info('Duration solved: %s' % duration)
sharpe_id_min = results_frame['sharpe'].idxmin()
max_sharpe_port = results_frame.iloc[sharpe_id_min]
stdev_id_min = results_frame['stdev'].idxmax()
min_vol_port = results_frame.iloc[stdev_id_min]
LOG.info('Max Sharpe ratio: %s' % str(max_sharpe_port))
LOG.info('Min standard deviation: %s' % str(min_vol_port))
db.connect()
solved = Portfolio()
max_item = parse_solved_portfolio(max_sharpe_port, stocks)
min_item = parse_solved_portfolio(min_vol_port, stocks)
solved.max_item = max_item
solved.min_item = min_item
solved.date = datetime.datetime.today()
solved.save()
db.close()
LOG.info('Save %d portfolio from %d in thread %s' %
(curr + 1, count, str(threading.get_ident())))
return 'Duration %s' % str(duration)
def parallel_solve_WTF_im_very_sad(all_stocks, type_ga, curr, count):
LOG.info('Start parallel solve in %s' % str(threading.get_ident()))
range_stock = len(all_stocks)
number = list()
stocks = []
for position in range(15):
pos = random.randint(0, range_stock)
while pos in number:
pos = random.randint(0, range_stock)
stock = all_stocks[pos - 1]
while len(stock.day_history) < 66:
pos = random.randint(0, range_stock)
stock = all_stocks[pos - 1]
number.append(pos)
stocks.append(stock)
cov_matrix, mean_daily_returns, arg_gmean, gmeans = cov_and_mean(stocks)
days = len(stocks[0].day_history)
iterations = 50000
start = time.time()
if type_ga == GA_SIMPLE:
results_frame = simple.solve(stocks, iterations, mean_daily_returns,
cov_matrix, days)
elif type_ga == GA_NSGAII:
results_frame = NSGAII.solve(stocks, iterations, mean_daily_returns,
cov_matrix, days)
else:
results_frame = NSGAII.solve_nsgaiii(stocks, iterations,
mean_daily_returns, cov_matrix,
days)
duration = time.time() - start
LOG.info('Duration solved: %s' % duration)
process_result_of_ga(results_frame, stocks, arg_gmean)
LOG.info('Save %d portfolio from %d in thread %s' %
(curr, count, str(threading.get_ident())))
return 'Duration %s' % str(duration)
start = time.time()
result_frame_sm = sm.solve(stocks, iter, mean_daily_returns, cov_matrix, days)
time_sm_solve = time.time() - start
time_sm.append(time_sm_solve)
LOG.info('sm-%d - time: %s' % (iter, str(time_sm_solve)))
start = time.time()
result_frame_nsgaii = nsagii.solve(stocks, iter, mean_daily_returns, cov_matrix, days,
generator=problemGenerator, population=1000)
time_nsga_solve = time.time() - start
time_nsga.append(time_nsga_solve)
LOG.info('nsgaii-%d - time: %s' % (iter, str(time_nsga_solve)))
start = time.time()
result_frame_nsgaiii = nsagii.solve_nsgaiii(stocks, iter, mean_daily_returns, cov_matrix, days,
generator=problemGenerator, population=1000)
time_nsga_solve = time.time() - start
time_nsgaiii.append(time_nsga_solve)
LOG.info('nsagiii-%d - time: %s' % (iter, str(time_nsga_solve)))
ax_1 = fig.add_subplot(1,3,1)
id_max = result_frame_sm['sharpe'].idxmax()
max_sharpe_port_sm = result_frame_sm.iloc[id_max]
id_max_wgmean = result_frame_sm['wgmean'].idxmax()
max_wgmean_port_sw = result_frame_sm.iloc[id_max_wgmean]
sharpes_sm.append(max_sharpe_port_sm[2])
wgmean_sm.append(max_wgmean_port_sw[1])
ax_1.scatter(result_frame_sm.stdev, result_frame_sm.ret, c=result_frame_sm.sharpe, cmap='RdYlBu')
ax_1.set_xlabel('stdev')
ax_1.set_ylabel('ret')
ax_1.scatter(max_sharpe_port_sm[2], max_sharpe_port_sm[0], marker=(3, 1, 0), color='r', s=300)