gemseo.algos.stop_criteria module#

Various termination criteria for drivers.

exception DesvarIsNan[source]#

Bases: TerminationCriterion

Stops driver when the design variables are nan.

exception FtolReached[source]#

Bases: TerminationCriterion

Exception raised when the f_tol_rel or f_tol_abs criteria is reached.

exception FunctionIsNan[source]#

Bases: TerminationCriterion

Stops driver when a function has NaN value or NaN Jacobian.

exception KKTReached[source]#

Bases: TerminationCriterion

A termination criterion based on the Karush-Kuhn-Tucker (KKT) residual norm.

exception MaxIterReachedException[source]#

Bases: TerminationCriterion

Exception raised when the maximum number of iterations is reached.

exception MaxTimeReached[source]#

Bases: TerminationCriterion

Exception raised when the maximum execution time is reached.

exception TerminationCriterion[source]#

Bases: Exception

Stop driver for some reason.

exception XtolReached[source]#

Bases: TerminationCriterion

Exception raised when the x_tol_rel or x_tol_abs criteria is reached.

class BaseToleranceTester(absolute=0.0, relative=0.0, n_last_iterations=3)[source]#

Bases: object

The base class to test the tolerance with respect to a reference value.

The reference value corresponds to the coordinate-wise average of the values associated to the last iterations.

Parameters:

  • absolute (float) --

    By default it is set to 0.0.

  • relative (float) --

    By default it is set to 0.0.

  • n_last_iterations (int) --

    By default it is set to 3.

check(problem, raise_exception=False, **kwargs)[source]#

Check whether the tolerance criterion is met.

Parameters:

  • problem (OptimizationProblem) -- The optimization problem to which the database is attached.

  • raise_exception (bool) --

    Whether to raise an exception when the tolerance criterion is not met.

    By default it is set to False.

  • **kwargs (Any) -- The options of the tester.

Returns:

Whether the tolerance criterion is not met.

Raises:

TerminationCriterion -- When the tolerance criterion is not met and raise_exception is True.

Return type:

bool

absolute: float = 0.0#

The absolute tolerance.

n_last_iterations: int = 3#

The number of last points to compute the reference.

relative: float = 0.0#

The relative tolerance.

termination_criterion: TerminationCriterion#

The termination criterion.

class DesignToleranceTester(absolute=0.0, relative=0.0, n_last_iterations=3)[source]#

Bases: BaseToleranceTester

A tolerance tester for the design_vector.

Parameters:

  • absolute (float) --

    By default it is set to 0.0.

  • relative (float) --

    By default it is set to 0.0.

  • n_last_iterations (int) --

    By default it is set to 3.

termination_criterion#

alias of XtolReached

class KKTConditionsTester(absolute=0.0, relative=0.0, n_last_iterations=3, ineq_tolerance=0.0, kkt_norm=0.0)[source]#

Bases: BaseToleranceTester

A tester for the Karush-Kuhn-Tucker (KKT) conditions.

Parameters:

  • absolute (float) --

    By default it is set to 0.0.

  • relative (float) --

    By default it is set to 0.0.

  • n_last_iterations (int) --

    By default it is set to 3.

  • ineq_tolerance (float) --

    By default it is set to 0.0.

  • kkt_norm (float) --

    By default it is set to 0.0.

termination_criterion#

alias of KKTReached

ineq_tolerance: float = 0.0#

The tolerance for the inequality constraints.

kkt_norm: float = 0.0#

The reference KKT norm.

class ObjectiveToleranceTester(absolute=0.0, relative=0.0, n_last_iterations=3)[source]#

Bases: BaseToleranceTester

A tolerance tester for the objective.

Parameters:

  • absolute (float) --

    By default it is set to 0.0.

  • relative (float) --

    By default it is set to 0.0.

  • n_last_iterations (int) --

    By default it is set to 3.

termination_criterion#

alias of FtolReached

is_f_tol_reached(opt_problem, f_tol_rel=1e-06, f_tol_abs=1e-06, n_x=2)[source]#

Tests if the tolerance on the objective function are reached.

The average function value of the last n_x points are taken Then it is checked that all points are within the distance of the center with relative and absolute tolerances specified by the user.

Parameters:

  • opt_problem (OptimizationProblem) -- the optimization problem containing the iterations

  • f_tol_rel (float) --

    relative tolerance

    By default it is set to 1e-06.

  • f_tol_abs (float) --

    absolute tolerance

    By default it is set to 1e-06.

  • n_x (int) --

    number of design vectors to account for

    By default it is set to 2.

Return type:

bool | bool_

is_kkt_residual_norm_reached(opt_problem, x_vect, kkt_abs_tol=0.0, kkt_rel_tol=0.0, ineq_tolerance=0.0001, reference_residual=1.0)[source]#

Test if the KKT conditions are satisfied.

Parameters:

  • opt_problem (OptimizationProblem) -- The optimization problem containing an optimization history.

  • x_vect (ndarray) -- The design point vector where the KKT conditions are tested.

  • kkt_abs_tol (float) --

    The absolute tolerance on the KKT condition residual.

    By default it is set to 0.0.

  • kkt_rel_tol (float) --

    The relative tolerance on the KKT condition residual.

    By default it is set to 0.0.

  • ineq_tolerance (float) --

    The tolerance to consider a constraint as active.

    By default it is set to 0.0001.

  • reference_residual (float) --

    The reference KKT condition residual.

    By default it is set to 1.0.

Returns:

Whether the absolute or the relative KKT residual norm criterion is reached.

Return type:

bool

is_x_tol_reached(opt_problem, x_tol_rel=1e-06, x_tol_abs=1e-06, n_x=2)[source]#

Tests if the tolerance on the design variables are reached.

The coordinate wise average of the last n_x points are taken Then it is checked that all points are within the distance of the center with relative and absolute tolerances specified by the user.

Parameters:

  • opt_problem (OptimizationProblem) -- the optimization problem containing the iterations

  • x_tol_rel (float) --

    relative tolerance

    By default it is set to 1e-06.

  • x_tol_abs (float) --

    absolute tolerance

    By default it is set to 1e-06.

  • n_x (int) --

    number of design vectors to account for

    By default it is set to 2.

Return type:

bool | bool_

kkt_residual_computation(opt_problem, x_vect, ineq_tolerance=0.0001)[source]#

Compute the KKT residual norm.

This implementation is inspired from Svanberg Matlab implementation of MMA algorithm see [Sva98]

Parameters:

  • opt_problem (OptimizationProblem) -- The optimization problem containing an optimization history.

  • x_vect (ndarray) -- The design point vector where the KKT conditions are tested.

  • ineq_tolerance (float) --

    The tolerance to consider a constraint as active.

    By default it is set to 0.0001.

Returns:

The KKT residual norm.

Return type:

float

KKT_RESIDUAL_NORM: Final[str] = 'KKT residual norm'#

The name to store the KKT residual norm in a database.