def __init__(self, pyopt_model, sens_type, **kwargs):
"""
:parameters:
:nlp: nonlinear problem pyOpt
:sens_type: sensitivity type
'FD' : estimation of gradients using finite differences
'CS' : estimation of gradients using complex step
grad_func : user provided gradients
"""
self.pyopt_model = pyopt_model
# Problem dimensions.
nbVar = len(pyopt_model._variables)
nbCons = len(pyopt_model._constraints)
# Bounds on variables.
LVar = -numpy.inf * numpy.ones(nbVar)
UVar = numpy.inf * numpy.ones(nbVar)
X0 = numpy.zeros(nbVar)
for i in range(nbVar):
var = pyopt_model.getVar(i)
LVar[i] = var.lower
UVar[i] = var.upper
X0[i] = var.value
# Constraints left and right-hand side.
LCon = -numpy.inf * numpy.ones(nbCons)
UCon = numpy.inf * numpy.ones(nbCons)
for j in range(nbCons):
cons = pyopt_model.getCon(j)
if cons.type=='i':
LCon[j] = cons.lower
UCon[j] = cons.upper
elif cons.type=='e':
LCon[j] = cons.equal
UCon[j] = cons.equal
# Differentiation method.
self.sens_type = sens_type
if sens_type == 'FD':
self.sens_step = 1e-6
else:
self.sens_step = 1e-20
self.gradient_method = Gradient(pyopt_model, sens_type, '',
self.sens_step)
# Saved values (private).
self._last_x = None
self._last_obj = None
self._last_grad_obj = None
self._last_cons = None
self._last_grad_con = None
# Initialize model.
NLPModel.__init__(self, name=pyopt_model.name, n=nbVar, m=nbCons, Lvar=LVar, Uvar=UVar,
Lcon=LCon, Ucon=UCon, x0=X0, **kwargs)
class NLPModel_From_PyOpt(NLPModel):
def __init__(self, pyopt_model, sens_type, **kwargs):
"""
:parameters:
:nlp: nonlinear problem pyOpt
:sens_type: sensitivity type
'FD' : estimation of gradients using finite differences
'CS' : estimation of gradients using complex step
grad_func : user provided gradients
"""
self.pyopt_model = pyopt_model
# Problem dimensions.
nbVar = len(pyopt_model._variables)
nbCons = len(pyopt_model._constraints)
# Bounds on variables.
LVar = -numpy.inf * numpy.ones(nbVar)
UVar = numpy.inf * numpy.ones(nbVar)
X0 = numpy.zeros(nbVar)
for i in range(nbVar):
var = pyopt_model.getVar(i)
LVar[i] = var.lower
UVar[i] = var.upper
X0[i] = var.value
# Constraints left and right-hand side.
LCon = -numpy.inf * numpy.ones(nbCons)
UCon = numpy.inf * numpy.ones(nbCons)
for j in range(nbCons):
cons = pyopt_model.getCon(j)
if cons.type=='i':
LCon[j] = cons.lower
UCon[j] = cons.upper
elif cons.type=='e':
LCon[j] = cons.equal
UCon[j] = cons.equal
# Differentiation method.
self.sens_type = sens_type
if sens_type == 'FD':
self.sens_step = 1e-6
else:
self.sens_step = 1e-20
self.gradient_method = Gradient(pyopt_model, sens_type, '',
self.sens_step)
# Saved values (private).
self._last_x = None
self._last_obj = None
self._last_grad_obj = None
self._last_cons = None
self._last_grad_con = None
# Initialize model.
NLPModel.__init__(self, name=pyopt_model.name, n=nbVar, m=nbCons, Lvar=LVar, Uvar=UVar,
Lcon=LCon, Ucon=UCon, x0=X0, **kwargs)
def obj(self, x):
if self._last_obj is not None and (self._last_x == x).all():
return self._last_obj
f, c, fail = self.pyopt_model.obj_fun(x)
self._last_x = x[:]
self._last_obj = f
self._last_cons = c
self._last_grad_obj = None # Gradient out of date.
self._last_grad_con = None # Gradient out of date.
return f
def cons(self, x):
if self._last_cons is not None and (self._last_x == x).all():
return self._last_cons
f, c, fail = self.pyopt_model.obj_fun(x)
self._last_x = x[:]
self._last_obj = f
self._last_cons = c
self._last_grad_obj = None # Gradient out of date.
self._last_grad_con = None # Gradient out of date.
return c
def grad(self, x):
if self._last_grad_obj is not None and (self._last_x == x).all():
return self._last_grad_obj
f, c, fail = self.pyopt_model.obj_fun(x)
grad = self.gradient_method.getGrad(x, {}, [f], c)
self._last_x = x[:]
self._last_obj = f
self._last_cons = c
self._last_grad_obj = grad[0].flatten()
self._last_grad_con = grad[1]
return grad[0].flatten()
def igrad(self, x, i):
if self._last_grad_con is not None and (self._last_x == x).all():
return self._last_grad_con[i]
f, c, fail = self.pyopt_model.obj_fun(x)
grad = self.gradient_method.getGrad(x, {}, [f], c)
self._last_x = x[:]
self._last_obj = f
self._last_cons = c
self._last_grad_obj = grad[0].flatten()
self._last_grad_con = grad[1]
return grad[1][i]