gemseo.algos.opt.core.trust_updater module#

Updates a trust parameter according to a decreases' ratio.

class BoundsUpdater(lower_bounds, upper_bounds, normalize=False)[source]#

Bases: object

Updater of the trust bounds, i.e. trust ball w.r.t.

the infinity norm.

Parameters:

  • lower_bounds (ndarray) -- The reference lower bounds.

  • upper_bounds (ndarray) -- The reference upper bounds.

  • normalize (bool) --

    Whether to apply the radius to the normalized bounds.

    By default it is set to False.

update(radius, center)[source]#

Update the trust bounds.

Parameters:

  • radius (float) -- The region radius w.r.t. the infinity norm.

  • center (ndarray) -- The center of the region

Returns:

The updated lower and upper bounds of the trust region.

Return type:

tuple[ndarray, ndarray]

class PenaltyUpdater(thresholds=(0.0, 0.25), multipliers=(0.5, 2.0), bound=1e-06)[source]#

Bases: TrustUpdater

Update the penalty parameter.

Parameters:

  • thresholds (tuple[float, float]) --

    The thresholds for the decreases' ratio.

    By default it is set to (0.0, 0.25).

  • multipliers (tuple[float, float]) --

    The multipliers for the trust parameter.

    By default it is set to (0.5, 2.0).

  • bound (float) --

    The absolute bound for the trust parameter.

    By default it is set to 1e-06.

update(ratio, parameter)[source]#

Update the trust parameter relative to the decrease ratio value.

Method to be overridden by subclasses.

Parameters:

  • ratio (float) -- The decrease ratio.

  • parameter (float) -- The trust parameter (radius or penalty).

Returns:

The new trust parameter, the iteration success.

Return type:

tuple[float, bool]

class RadiusUpdater(thresholds=(0.0, 0.25), multipliers=(0.5, 2.0), bound=1000.0)[source]#

Bases: TrustUpdater

Update the radius of the trust region.

Parameters:

  • thresholds (tuple[float, float]) --

    The thresholds for the decreases' ratio.

    By default it is set to (0.0, 0.25).

  • multipliers (tuple[float, float]) --

    The multipliers for the trust parameter.

    By default it is set to (0.5, 2.0).

  • bound (float) --

    The absolute bound for the trust parameter.

    By default it is set to 1000.0.

update(ratio, parameter)[source]#

Update the trust parameter relative to the decrease ratio value.

Method to be overridden by subclasses.

Parameters:

  • ratio (float) -- The decrease ratio.

  • parameter (float) -- The trust parameter (radius or penalty).

Returns:

The new trust parameter, the iteration success.

Return type:

tuple[float, bool]

class TrustUpdater(thresholds, multipliers, bound=None)[source]#

Bases: object

Updater of a trust parameter.

Parameters:

  • thresholds (tuple[float, float]) -- The thresholds for the decreases' ratio.

  • multipliers (tuple[float, float]) -- The multipliers for the trust parameter.

  • bound -- The absolute bound for the trust parameter.

update(ratio, parameter)[source]#

Update the trust parameter relative to the decrease ratio value.

Method to be overridden by subclasses.

Parameters:

  • ratio (float) -- The decrease ratio.

  • parameter (float) -- The trust parameter (radius or penalty).

Returns:

The new trust parameter, the iteration success.

Return type:

tuple[float, bool]