def optimize(p0, fun, method='BFGS', args=None, jac=None, callback='default', options=None):
glob_counter = 0
def _callback_default(xi):
global glob_counter
if glob_counter % 10 == 0:
count = str(glob_counter)
while len(count)<4:count=' '+count
print 'iter:', count, "%0.5e" % fun(xi, *args)
pi = utils.vector_2_para(xi, p0)
glob_counter += 1
def _callback_none(xi):pass
if callback == None:
_callback = _callback_none
if callback == 'default':
_callback = _callback_default
if callback is not 'default' and callback is not None:
table, opt = callback(p0)
table = collections.OrderedDict(sorted(table.items()))
res = ''
for item in table:
while len(item)<11:item=item+' '
res+=item[:10]+' '
print 'Epoch: ' + res
print '-'*(len(table)*11+11)
def _callback(xi):
global glob_counter
glob_counter += 1
if glob_counter==1 or glob_counter%opt['freq']==0 or glob_counter==options['maxiter']:
pi = utils.vector_2_para(xi,p0)
table, _ = callback(pi)
table = collections.OrderedDict(sorted(table.items()))
line = np.array([glob_counter]+table.values())
for i in line:
s = str('%.4g' % i)
while len(s)<9:s=s+' '
print s+' ',
print
res = scipy.optimize.minimize(
x0 = utils.para_2_vector(p0),
fun = fun,
method = method,
callback = _callback,
args = args,
jac = jac,
options = options
)
p1 = deepcopy( utils.vector_2_para(res['x'], p0) )
return p1, res
def exe(fun, args):
para = args[0]
rest = args[1:]
v = utils.para_2_vector(para)
return fun(v, *rest)