class documentation

class ConstrainedFitnessAL:

View In Hierarchy

Construct an unconstrained objective function from constraints.

This class constructs an unconstrained "fitness" function (to be minimized) from an objective function and an inequality constraints function (which returns a list of constraint values). An equality constraint h(x) == 0 must be expressed as two inequality constraints like [h(x) - eps, -h(x) - eps] with eps >= 0. Non-positive values <= 0 are considered feasible.

The update method of the class instance needs to be called after each iteration. Depending on the setting of which, update may call get_solution(es) which shall return the solution to be used for the constraints handling update, by default _get_favorite_solution == lambda es: es.ask(1, sigma_fac=0)[0]. The additional evaluation of objective and constraints is avoided by the default which='best', using the best solution in the current iteration.

find_feasible_first optimizes to get a feasible solution first, which only works if no equality constraints are implemented. For this reason the default is False.

Minimal example (verbosity set for doctesting):

>>> import cma
>>> def constraints(x):  # define the constraint
...     return [x[0] + 1, x[1]]  # shall be <= 0
>>> cfun = cma.ConstrainedFitnessAL(cma.ff.sphere, constraints,
...                                 find_feasible_first=True)
>>> es = cma.CMAEvolutionStrategy(3 * [1.1], 0.1,
...                   {'tolstagnation': 0, 'verbose':-9})  # verbosity for doctest only
>>> es = es.optimize(cfun, callback=cfun.update)
>>> x = es.result.xfavorite

The best x return value of cma.fmin2 may not be useful, because the underlying function changes over time. Therefore, we use es.result.xfavorite, which is still not guarantied to be a feasible solution. Alternatively, cfun.best_feas.x contains the best evaluated feasible solution. However, this is not necessarily expected to be a good solution, see below.

>>> assert sum((x - [-1, 0, 0])**2) < 1e-9, x
>>> assert es.countevals < 2200, es.countevals
>>> assert cfun.best_feas.f < 10, str(cfun.best_feas)
>>> # print(es.countevals, cfun.best_feas.__dict__)

To find a final feasible solution (close to es.result.xfavorite) we can use the current CMAEvolutionStrategy instance es:

>>> x = cfun.find_feasible(es)  # uses es.optimize to find (another) feasible solution
>>> assert constraints(x)[0] <= 0, (x, cfun.best_feas.x)
>>> assert cfun.best_feas.f < 1 + 2e-6, str(cfun.best_feas)
>>> assert len(cfun.archives) == 3
>>> assert cma.ConstrainedFitnessAL(cma.ff.sphere, constraints, archives=False).archives == []

Details: The fitness, to be minimized, is changing over time such that the overall minimal value does not indicate the best solution.

The construction is based on the AugmentedLagrangian class. If, as by default, self.finding_feasible is False, the fitness equals f(x) + sum_i (lam_i x g_i + mu_i x g_i / 2) where g_i = max(g_i(x), -lam_i / mu_i) and lam_i and mu_i are generally positive and dynamically adapted coefficients. Only lam_i can change the position of the optimum in the feasible domain (and hence must converge to the right value).

When self.finding_feasible is True, the fitness equals to sum_i (g_i > 0) x g_i^2 and omits f + sum_i lam_i g_i altogether. Whenever a feasible solution is found, the finding_feasible flag is reset to False.

find_feasible(es) sets finding_feasible = True and uses es.optimize to optimize self.__call__. This works well with CMAEvolutionStrategy but may easily fail with solvers that do not consistently pass over the optimum in search space but approach the optimum from one side only. This is not advisable if the feasible domain has zero volume, e.g. when g models an equality like g = lambda x: [h(x), -h(x)].

An equality constraint, h(x) = 0, cannot be handled like h**2 <= 0, because the Augmented Lagrangian requires the derivative at h == 0 to be non-zero. Using abs(h) <= 0 leads to divergence of coefficient mu and the condition number. The best way is apparently using the two inequality constraints [h <= 0, -h <= 0], which seems to work perfectly well. The underlying AugmentedLagrangian class also accepts equality constraints.

Method __call__ return AL fitness, append f and g values to self.F and self.G.
Method __init__ constructor with lazy initialization.
Method find​_feasible find feasible solution by calling es.optimize(self).
Method initialize set search space dimension explicitely
Method log​_in​_es a hack to have something in the cma-logger divers plot.
Method reset reset dynamic components
Method set​_mu​_lam set AL coefficients
Method update update AL coefficients, may be used as callback to OOOptimizer.optimize.
Class Variable archive​_aggregators Undocumented
Instance Variable archives Undocumented
Instance Variable best​_aug Undocumented
Instance Variable best​_f​_plus​_gpos Undocumented
Instance Variable best​_feas Undocumented
Instance Variable constraints Undocumented
Instance Variable count​_calls Undocumented
Instance Variable count​_updates Undocumented
Instance Variable dimension Undocumented
Instance Variable F Undocumented
Instance Variable ​F_plus_sum_al_​G Undocumented
Instance Variable find​_feasible​_aggregator Undocumented
Instance Variable finding​_feasible Undocumented
Instance Variable foffset Undocumented
Instance Variable fun Undocumented
Instance Variable G Undocumented
Instance Variable get​_solution Undocumented
Instance Variable logging Undocumented
Instance Variable omit​_f​_calls​_when​_possible Undocumented
Instance Variable which Undocumented
Property al AugmentedLagrangian class instance
Method ​_fg​_values f, g values used to update the Augmented Lagrangian coefficients
Method ​_is​_feasible return True if last evaluated solution (or gvals) was feasible
Method ​_reset Undocumented
Method ​_reset​_archives Undocumented
Method ​_reset​_arrays Undocumented
Method ​_update​_best keep track of best solution and best feasible solution
Instance Variable ​_al Undocumented
Instance Variable ​_set​_coefficient​_counts Undocumented
Property ​_best​_fg return current best f, g, where best is determined by the Augmented Lagrangian
def __call__(self, x):

return AL fitness, append f and g values to self.F and self.G.

If self.finding_feasible, fun(x) is not called and f = np.nan.

def __init__(self, fun, constraints, dimension=None, which='best', find_feasible_first=False, get_solution=_get_favorite_solution, logging=None, archives=archive_aggregators):

constructor with lazy initialization.

If which in ['mean', 'solution'], get_solution is called (with the argument passed to the update method) to determine the solution used to update the AL coefficients.

If find_feasible_first, only the constraints are optimized until the first (fully) feasible solution is found.

logging is the iteration gap for logging constraints related data, in AugmentedLagrangian. 0 means no logging and negative values have unspecified behavior.

archives are the aggregator functions for constraints for non-dominated biobjective archives. By default, the second objective is max(g_+), sum(g_+) or sum(g_+^2), respectively. archives=True invokes the same behavior. archives=False or an empty tupel prevents maintaining archives.

def find_feasible(self, es, termination=('maxiter', 'maxfevals'), aggregator=None):

find feasible solution by calling es.optimize(self).

Return best ever feasible solution self.best_feas.x. See also self.best_feas.info.

aggregator, defaulting to self.find_feasible_aggregator, is the constraints aggregation function used as objective function to be minimized. aggregator takes as input all constraint values and returns a value <= 0 if and only if the solution is feasible.

Terminate when either (another) feasible solution was found or any of the termination keys is matched in es.stop().

def initialize(self, dimension):
set search space dimension explicitely
def log_in_es(self, es, f, g):

a hack to have something in the cma-logger divers plot.

Append the sum of positive g-values and the number of infeasible constraints, displayed like 10**(number/10) (mapping [0, 10] to [1, 10]) if number < 10, to es.more_to_write.

def reset(self):
reset dynamic components
def set_mu_lam(self, mu, lam):
set AL coefficients
def update(self, es):

update AL coefficients, may be used as callback to OOOptimizer.optimize.

TODO: decide what happens when __call__ was never called:
ignore (as for now) or update based on xfavorite by calling self(xfavorite), assuming that update was called on purpose? When method is not best, it should work without even call self(xfavorite).
archive_aggregators =

Undocumented

archives =

Undocumented

best_aug =

Undocumented

best_f_plus_gpos =

Undocumented

best_feas =

Undocumented

constraints =

Undocumented

count_calls: int =

Undocumented

count_updates: int =

Undocumented

dimension =

Undocumented

F =

Undocumented

F_plus_sum_al_G =

Undocumented

find_feasible_aggregator =

Undocumented

finding_feasible: bool =

Undocumented

foffset =

Undocumented

fun =

Undocumented

G =

Undocumented

get_solution =

Undocumented

logging =

Undocumented

omit_f_calls_when_possible: bool =

Undocumented

which =

Undocumented

@property
al =
AugmentedLagrangian class instance
def _fg_values(self, es):
f, g values used to update the Augmented Lagrangian coefficients
def _is_feasible(self, gvals=None):
return True if last evaluated solution (or gvals) was feasible
def _reset(self):

Undocumented

def _reset_archives(self):

Undocumented

def _reset_arrays(self):

Undocumented

def _update_best(self, x, f, g, g_al=None):
keep track of best solution and best feasible solution
_al =

Undocumented

_set_coefficient_counts: list =

Undocumented

@property
_best_fg =
return current best f, g, where best is determined by the Augmented Lagrangian