import numpy as np
from iwopy.core import SingleObjOptResults, MultiObjOptResults
from . import imports
[docs]class SingleObjProblemTemplate:
"""
Template for a wrapper around the pymoo problem
for a single objective.
At the moment this interface only supports
pure int or pure float problems (not mixed).
Attributes
----------
problem: iwopy.core.Problem
The iwopy problem to solve
vectorize: bool
Switch for vectorized calculations, wrt
population individuals
is_mixed: bool
Flag for mixed integer/float problems
is_intprob: bool
Flag for integer problems
:group: interfaces.pymoo
"""
[docs] def __init__(self, problem, vectorize):
"""
Constructor
Parameters
----------
problem: iwopy.core.Problem
The iwopy problem to solve
vectorize: bool, optional
Switch for vectorized calculations, wrt
population individuals
"""
self.problem = problem
self.vectorize = vectorize
if self.problem.n_vars_float > 0 and self.problem.n_vars_int == 0:
self.is_mixed = False
self.is_intprob = False
self._pargs = dict(
n_var=self.problem.n_vars_float,
n_obj=self.problem.n_objectives,
n_ieq_constr=self.problem.n_constraints,
xl=self.problem.min_values_float(),
xu=self.problem.max_values_float(),
elementwise=not vectorize,
type_var=np.float64,
)
elif self.problem.n_vars_float == 0 and self.problem.n_vars_int > 0:
self.is_mixed = False
self.is_intprob = True
self._pargs = dict(
n_var=self.problem.n_vars_int,
n_obj=self.problem.n_objectives,
n_ieq_constr=self.problem.n_constraints,
xl=self.problem.min_values_int(),
xu=self.problem.max_values_int(),
elementwise=not vectorize,
type_var=np.int32,
)
else:
self.is_mixed = True
self.is_intprob = False
vars = {}
nami = self.problem.var_names_int()
inii = self.problem.initial_values_int()
mini = self.problem.min_values_int()
maxi = self.problem.max_values_int()
for i, v in enumerate(nami):
vars[v] = imports.Integer(value=inii[i], bounds=(mini[i], maxi[i]))
namf = self.problem.var_names_float()
inif = self.problem.initial_values_float()
minf = self.problem.min_values_float()
maxf = self.problem.max_values_float()
for i, v in enumerate(namf):
vars[v] = imports.Real(value=inif[i], bounds=(minf[i], maxf[i]))
self._pargs = dict(
vars=vars,
n_obj=self.problem.n_objectives,
n_ieq_constr=self.problem.n_constraints,
elementwise=not vectorize,
)
if self.problem.n_constraints:
self._cmi = self.problem.min_values_constraints
self._cma = self.problem.max_values_constraints
cnames = self.problem.cons.component_names
sel = np.isinf(self._cmi) & np.isinf(self._cma)
if np.any(sel):
raise RuntimeError(
f"Missing boundaries for constraints {cnames[sel]}"
)
sel = (~np.isinf(self._cmi)) & (~np.isinf(self._cma))
if np.any(sel):
raise RuntimeError(
f"Constraints {cnames[sel]} have both lower and upper bounds"
)
def _evaluate(self, x, out, *args, **kwargs):
"""
Overloading the abstract evaluation function
of the pymoo base class.
"""
# vectorized run:
if self.vectorize:
if self.is_mixed:
xi = np.array(
[[dct[v] for v in self.problem.var_names_int()] for dct in x],
dtype=np.int32,
)
xf = np.array(
[[dct[v] for v in self.problem.var_names_float()] for dct in x],
dtype=np.float64,
)
out["F"], out["G"] = self.problem.evaluate_population(xi, xf)
out["F"] *= np.where(self.problem.maximize_objs, -1.0, 1.0)[None, :]
else:
n_pop = x.shape[0]
if self.is_intprob:
dummies = np.zeros((n_pop, 0), dtype=np.float64)
out["F"], out["G"] = self.problem.evaluate_population(
x, dummies
)
out["F"] *= np.where(self.problem.maximize_objs, -1.0, 1.0)[
None, :
]
else:
dummies = np.zeros((n_pop, 0), dtype=np.int32)
out["F"], out["G"] = self.problem.evaluate_population(
dummies, x
)
out["F"] *= np.where(self.problem.maximize_objs, -1.0, 1.0)[
None, :
]
if self.problem.n_constraints:
sel = ~np.isinf(self._cma)
out["G"][:, sel] = out["G"][:, sel] - self._cma[None, sel]
sel = ~np.isinf(self._cmi)
out["G"][:, sel] = self._cmi[None, sel] - out["G"][:, sel]
# individual run:
else:
if self.is_mixed:
xi = np.array(
[x[v] for v in self.problem.var_names_int()], dtype=np.int32
)
xf = np.array(
[x[v] for v in self.problem.var_names_float()], dtype=np.float64
)
out["F"], out["G"] = self.problem.evaluate_individual(xi, xf)
out["F"] *= np.where(self.problem.maximize_objs, -1.0, 1.0)
else:
n_pop = x.shape[0]
if self.is_intprob:
dummies = np.zeros(0, dtype=np.float64)
out["F"], out["G"] = self.problem.evaluate_individual(
x, dummies
)
out["F"] *= np.where(self.problem.maximize_objs, -1.0, 1.0)
else:
dummies = np.zeros(0, dtype=np.int32)
out["F"], out["G"] = self.problem.evaluate_individual(
dummies, x
)
out["F"] *= np.where(self.problem.maximize_objs, -1.0, 1.0)
if self.problem.n_constraints:
sel = ~np.isinf(self._cma)
out["G"][sel] = out["G"][sel] - self._cma[sel]
sel = ~np.isinf(self._cmi)
out["G"][sel] = self._cmi[sel] - out["G"][sel]
[docs] def finalize(self, pymoo_results, verbosity=1):
"""
Finalize the problem.
Parameters
----------
pymoo_results: pymoo.Results
The results from the solver
verbosity: int
The verbosity level, 0 = silent
Returns
-------
results: iwopy.SingleObjOptResults
The optimization results object
"""
# prepare:
r = pymoo_results
suc = True
# case no solution from pymoo:
if r.X is None:
suc = False
xi = None
xf = None
res = None
objs = None
cons = None
# evaluate pymoo final solution:
else:
if self.is_mixed:
xi = np.array(
[r.X[v] for v in self.problem.var_names_int()], dtype=np.int32
)
xf = np.array(
[r.X[v] for v in self.problem.var_names_float()],
dtype=np.float64,
)
else:
if self.is_intprob:
xi = r.X
xf = np.zeros(0, dtype=np.float64)
else:
xi = np.zeros(0, dtype=np.int32)
xf = np.array(r.X, dtype=np.float64)
if self.vectorize:
if self.is_mixed:
pxi = np.array(
[
[p.X[v] for v in self.problem.var_names_int()]
for p in r.pop
],
dtype=np.int32,
)
pxf = np.array(
[
[p.X[v] for v in self.problem.var_names_float()]
for p in r.pop
],
dtype=np.float64,
)
self.problem.finalize_population(pxi, pxf, verbosity)
del pxi, pxf
else:
n_pop = len(r.pop)
n_vars = len(r.X)
vars = np.zeros((n_pop, n_vars), dtype=np.float64)
for pi, p in enumerate(r.pop):
vars[pi] = p.X
if self.is_intprob:
dummies = np.zeros((n_pop, 0), dtype=np.int32)
self.problem.finalize_population(
vars.astype(np.int32), dummies, verbosity
)
else:
dummies = np.zeros((n_pop, 0), dtype=np.float64)
self.problem.finalize_population(dummies, vars, verbosity)
res, objs, cons = self.problem.finalize_individual(xi, xf, verbosity)
if verbosity:
print()
suc = np.all(self.problem.check_constraints_individual(cons, False))
if verbosity:
print()
return SingleObjOptResults(self.problem, suc, xi, xf, objs, cons, res)
[docs] @classmethod
def get_class(cls):
"""
Creates the class, dynamically derived from pymoo.Problem
"""
imports.load()
attrb = {v: d for v, d in cls.__dict__.items()}
init0 = cls.__init__
def init1(self, *args, **kwargs):
init0(self, *args, **kwargs)
imports.Problem.__init__(self, **self._pargs)
attrb["__init__"] = init1
attrb["__doc__"] = attrb["__doc__"].replace("Template for a w", "W")
del attrb["get_class"]
return type("SingleObjProblem", (imports.Problem,), attrb)
[docs]class MultiObjProblemTemplate:
"""
Template for a wrapper around the pymoo problem
for a multiple objectives problem.
At the moment this interface only supports
pure int or pure float problems (not mixed).
Attributes
----------
problem: iwopy.core.Problem
The iwopy problem to solve
vectorize: bool
Switch for vectorized calculations, wrt
population individuals
is_intprob: bool
Flag for integer problems
:group: interfaces.pymoo
"""
[docs] def __init__(self, problem, vectorize):
"""
Constructor template, will be overwritten by get_class
Parameters
----------
problem: iwopy.core.Problem
The iwopy problem to solve
vectorize: bool, optional
Switch for vectorized calculations, wrt
population individuals
"""
pass
[docs] def finalize(self, pymoo_results, verbosity=1):
"""
Finalize the problem.
Parameters
----------
pymoo_results: pymoo.Results
The results from the solver
verbosity: int
The verbosity level, 0 = silent
Returns
-------
results: iwopy.SingleObjOptResults
The optimization results object
"""
# prepare:
r = pymoo_results
suc = True
# case no solution from pymoo:
if r.X is None:
suc = False
xi = None
xf = None
res = None
objs = None
cons = None
# evaluate pymoo final solution:
else:
n_pop = len(r.pop)
if self.is_intprob:
xi = r.X
xf = np.zeros((n_pop, 0), dtype=np.float64)
else:
xi = np.zeros((n_pop, 0), dtype=int)
xf = r.X
res, objs, cons = self.problem.finalize_population(xi, xf, verbosity)
if verbosity:
print()
suc = np.all(
self.problem.check_constraints_population(cons, False), axis=1
)
if verbosity:
print()
return MultiObjOptResults(self.problem, suc, xi, xf, objs, cons, res)
[docs] @classmethod
def get_class(cls):
"""
Creates the class, dynamically derived from SingleObjProblem
"""
scls = SingleObjProblemTemplate.get_class()
attrb = {v: d for v, d in cls.__dict__.items()}
def init(self, *args, **kwargs):
scls.__init__(self, *args, **kwargs)
attrb["__init__"] = init
attrb["__doc__"] = attrb["__doc__"].replace("Template for a w", "W")
del attrb["get_class"]
return type("MultiObjProblem", (scls,), attrb)