def make_optimiser(config, model):
if not model: return None
# run with dummy values -- this doesn't really belong here, but temporarily...
if config['wetrun']:
print "--\nGenerating optimisable function"
ff = model.optimisable(config['distance'], debug=config['debug'])
print ff
print "--\nAttempting single run of function"
pp = [x.get('default', 0) for x in config['params']]
ff(pp)
return None
mode = config.get('job_mode', JOB_MODE)
weights = [config['weights'].get(x['name'], 1) for x in config['vars']]
if mode == 'GLP':
lb = [x.get('min', 0) for x in config['params']]
ub = [x.get('max', 0) for x in config['params']]
return openopt.GLP(
model.optimisable(config['distance'], weights=weights),
lb=lb,
ub=ub,
# TODO, possibly: support A and b args to define linear constraints
maxIter=config['max_iter'])
elif mode == 'NLP':
lb = [x.get('min', 0) for x in config['params']]
ub = [x.get('max', 0) for x in config['params']]
return openopt.NLP(
model.optimisable(config['distance'], weights=weights),
lb=lb,
ub=ub,
# TODO, possibly: support A and b args to define linear constraints
maxIter=config['max_iter'])
elif mode == 'NSP':
lb = [x.get('min', 0) for x in config['params']]
ub = [x.get('max', 0) for x in config['params']]
return openopt.NSP(
model.optimisable(config['distance'], weights=weights),
lb=lb,
ub=ub,
# TODO, possibly: support A and b args to define linear constraints
maxIter=config['max_iter'])
return None
"penalty: {}".format(penalty)
print
return sharpe - penalty
if args.weights is None:
initial_weights = np.ones(len(forecasts)) * .5
else:
initial_weights = np.array([float(x) for x in args.weights.split(",")])
lb = np.ones(len(forecasts)) * 0.0
ub = np.ones(len(forecasts))
plotit = False
p = openopt.NSP(goal='max',
f=objective,
x0=initial_weights,
lb=lb,
ub=ub,
plot=plotit)
p.ftol = 0.001
p.maxFunEvals = 150
r = p.solve('ralg')
if (r.stopcase == -1 or r.isFeasible == False):
print
objective_detail(target, *g_params)
raise Exception("Optimization failed")
print
r.xf
ii = 0
for fcast in forecasts:
return (pret * 252) / np.sqrt(pvar * 252)
mean = 0
cnt = 0
gstart = pd.to_datetime("20110101")
start = pd.to_datetime("20110101")
end = pd.to_datetime("20110101") + timedelta(days=30)
while end < pd.to_datetime("20130101"):
lb = np.ones(10) * 0.0
ub = np.ones(10)
plotit = False
initial_weights = np.asarray([.5, .5, .5, .5, .5, .5, .5, .5, .5, .5])
#initial_weights = np.asarray([0, 0, 0, 0, 1, 0, 0, 0, 0, 0])
p = openopt.NSP(goal='max', f=fcn, x0=initial_weights, lb=lb, ub=ub)
p.args.f = (start, end)
p.ftol = 0.001
p.maxFunEvals = 300
r = p.solve('ralg')
if (r.stopcase == -1 or r.isFeasible == False):
print(objective_detail(target, *g_params))
raise Exception("Optimization failed")
print(r.xf)
for ii in range(0, 10):
print("{}: {}".format(cols[ii], r.xf[ii]))
ii += 1
wtrecent = r.xf