def minimize_f(fi, K = None, method = None):
"""
Minimize the ``fi`` function instance. Returns the number of minimization
iterations performed and method finding the optimum (if any).
"""
f = fi.f
optmethod = None
pop = Population(fi, K, [XMinimizeMethod(name, fi) for name in string.split(method, ',')])
# Iterate; make a full iteration even in case maxfunevals is reached.
stop = False
while not stop:
for i in range(K):
(x, y) = pop.step_one(i)
#print("[%d] #%d %s=%s" % (pop.total_iters, i, x, y))
if y < f.ftarget:
optmethod = pop.minimizers[i].minmethod.name
stop = True
break # stop immediately, no point in going on
if f.evaluations > fi.maxfunevals:
stop = True
pop.end_iter()
pop.stop()
return (pop.total_iters, optmethod)
def minimize_f(fi, K=None, method=None):
"""
Minimize the ``fi`` function instance. Returns the number of minimization
iterations performed and method finding the optimum (if any).
"""
f = fi.f
optmethod = None
pop = Population(
fi, K,
[XMinimizeMethod(name, fi) for name in string.split(method, ',')])
# Iterate; make a full iteration even in case maxfunevals is reached.
stop = False
while not stop:
for i in range(K):
(x, y) = pop.step_one(i)
#print("[%d] #%d %s=%s" % (pop.total_iters, i, x, y))
if y < f.ftarget:
optmethod = pop.minimizers[i].minmethod.name
stop = True
break # stop immediately, no point in going on
if f.evaluations > fi.maxfunevals:
stop = True
pop.end_iter()
pop.stop()
return (pop.total_iters, optmethod)
def minimize_f(fi, K=None, method=None, eps=None, accrual=None):
"""
Minimize the ``fi`` function instance. Returns the number of minimization
iterations performed and method finding the optimum (if any).
"""
f = fi.f
optmethod = None
pop = Population(
fi, K,
[MinimizeMethod(name, fi) for name in string.split(method, ',')])
# Credit Assignment does not matter for us as long as it's order-invariant
popcredit = PopulationCredit(pop, "raw", accrual)
stop = False
# Initial iterations - evaluate each function once.
for i in range(K):
(x, y) = pop.step_one(i)
#print("[%d] #%d %s=%s" % (pop.total_iters, i, x, y))
if y < f.ftarget:
optmethod = pop.minimizers[i].minmethod.name
stop = True
break # stop immediately, no point in going on
pop.end_iter()
popcredit.update()
# Main set of iterations - explore/exploit.
while not stop and f.evaluations < fi.maxfunevals:
if np.random.rand() < eps:
# Explore
i = np.random.randint(K)
else:
# Exploit
i = popcredit.credit.argmin()
(x, y) = pop.step_one(i)
#print("[%d] #%d %s=%s" % (pop.total_iters, i, x, y))
if y < f.ftarget:
optmethod = pop.minimizers[i].minmethod.name
stop = True
break # stop immediately, no point in going on
pop.end_iter()
popcredit.update()
pop.stop()
return (pop.total_iters, optmethod)
def minimize_f(fi, K = None, method = None, eps = None, accrual = None):
"""
Minimize the ``fi`` function instance. Returns the number of minimization
iterations performed and method finding the optimum (if any).
"""
f = fi.f
optmethod = None
pop = Population(fi, K, [MinimizeMethod(name, fi) for name in string.split(method, ',')])
# Credit Assignment does not matter for us as long as it's order-invariant
popcredit = PopulationCredit(pop, "raw", accrual)
stop = False
# Initial iterations - evaluate each function once.
for i in range(K):
(x, y) = pop.step_one(i)
#print("[%d] #%d %s=%s" % (pop.total_iters, i, x, y))
if y < f.ftarget:
optmethod = pop.minimizers[i].minmethod.name
stop = True
break # stop immediately, no point in going on
pop.end_iter()
popcredit.update()
# Main set of iterations - explore/exploit.
while not stop and f.evaluations < fi.maxfunevals:
if np.random.rand() < eps:
# Explore
i = np.random.randint(K)
else:
# Exploit
i = popcredit.credit.argmin()
(x, y) = pop.step_one(i)
#print("[%d] #%d %s=%s" % (pop.total_iters, i, x, y))
if y < f.ftarget:
optmethod = pop.minimizers[i].minmethod.name
stop = True
break # stop immediately, no point in going on
pop.end_iter()
popcredit.update()
pop.stop()
return (pop.total_iters, optmethod)