def infer_x(self, y, sigma = 1.0, tolerance = 1e-2):
"""Infer probable x from input y
@param y the desired output for infered x.
@return a list of probable x
"""
OptimizedInverseModel.infer_x(self, y)
if self.fmodel.size() == 0:
return self._random_x()
x_guesses = [self._guess_x_simple(y)[0]]
#x_guesses = self._guess_x_kmeans(y)
if CMA_IMPLEM == "C++":
# C++ IMPLEM
result = []
for xg in x_guesses:
sigma = 0.05
#print self.lower, self.upper
p = lci.to_params(list(xg),
sigma,
lbounds=self.lower,
ubounds=self.upper,
max_fevals=20
)
objfunc = lci.to_fitfunc(self._error)
cmasols = lci.pcmaes(objfunc, p)
# collect and inspect results
bcand = cmasols.best_candidate()
error = bcand.get_fvalue()
x = lcmaes.get_candidate_x(bcand)
result.append((error, x))
#print "x", [xi for fi, xi in sorted(result)]
return [xi for fi, xi in sorted(result)]
elif CMA_IMPLEM == "python":
# PYTHON IMPLEM
result = []
for xg in x_guesses:
res = cma.fmin(self._error, xg, 0.05,
options={'bounds':[self.lower, self.upper],
'verb_log':0,
'verb_disp':False,
#'ftarget':tolerance/2,
'maxfevals':20})
result.append((res[1], res[0]))
# return [xi for fi, i, xi in sorted(result)[:min(N_RESULT, len(result))]]
return [xi for fi, xi in sorted(result)]
def fit(self, X, y, opt_fit=False, libcma=False, **params):
if opt_fit:
def err_function(params):
mdl = NrmModel(bias=params[0],
conf_threshold=params[1],
mu0=params[2],
kappa0=params[3],
alpha0=params[4],
beta0=params[5])
return -mdl.score(X, y)
start = np.array([self.bias, self.conf_threshold] +
list(self.prior))
if not libcma:
import cma
x = cma.fmin(err_function,
start,
1.5,
restarts=5,
options={'verbose': -9})[0]
else:
import lcmaes
import lcmaes_interface as lci
print(('Start:', err_function(start)))
ffunc = lci.to_fitfunc(err_function)
lbounds = [-np.inf, 0, -np.inf] + 3 * [0]
fopt = lci.to_params(
list(start),
1.5,
str_algo=b'aipop',
lbounds=lbounds,
restarts=5,
)
res = lci.pcmaes(ffunc, fopt)
bcand = res.best_candidate()
print(('End:', bcand.get_fvalue()))
x = lcmaes.get_candidate_x(bcand)
return self.set_params(bias=x[0],
conf_threshold=x[1],
mu0=x[2],
kappa0=x[3],
alpha0=x[4],
beta0=x[5])
return self
import lcmaes_interface as lci
# setup input parameters
def myfun(x):
return sum([xi**2 for xi in x]) # myfun accepts a list of numbers as input
dimension = 10
x0 = [2.1] * dimension
sigma0 = 0.1
# run optimization via lci
res = lci.pcmaes(
lci.to_fitfunc(myfun),
lci.to_params(
x0,
sigma0, # all remaining parameters are optional
str_algo=b'aipop', # b=bytes, because unicode fails
lbounds=[-5] * dimension,
ubounds=[5] * dimension,
restarts=2, # means 2 runs, i.e. one restart
))
lci.plot() # plot from file set in lci.to_params
import lcmaes_interface as lci
# setup input parameters
def myfun(x): return sum([xi**2 for xi in x]) # myfun accepts a list of numbers as input
dimension = 10
x0 = [2.1] * dimension
sigma0 = 0.1
# run optimization via lci
res = lci.pcmaes(lci.to_fitfunc(myfun),
lci.to_params(x0, sigma0, # all remaining parameters are optional
str_algo=b'aipop', # b=bytes, because unicode fails
lbounds=[-5] * dimension, ubounds=[5] * dimension,
restarts=2, # means 2 runs, i.e. one restart
)
)
lci.plot() # plot from file set in lci.to_params