Termination#

Layer 2: Compose

`saealib.Termination`	Termination class to determine when to stop the optimization process.
`saealib.TerminationCondition`	Composable termination condition.

saealib.max_fe(value)[source]#

Create a termination condition based on maximum function evaluations.

Parameters:: value (int) – Maximum number of function evaluations.
Returns:: A composable condition that returns True when ctx.fe >= value.
Return type:: TerminationCondition

Examples

>>> termination = Termination(max_fe(2000))
>>> condition = max_fe(2000) & max_gen(100)

saealib.max_gen(value)[source]#

Create a termination condition based on maximum generations.

Parameters:: value (int) – Maximum number of generations.
Returns:: A composable condition that returns True when ctx.gen >= value.
Return type:: TerminationCondition

Examples

>>> termination = Termination(max_gen(100))
>>> condition = max_gen(100) | max_fe(2000)

saealib.f_target(value)[source]#

Create a termination condition based on a target objective value.

Intended for single-objective problems. The direction is taken from ctx.direction (-1 for minimization, +1 for maximization), so the condition is met when the best objective found reaches value from the correct side (best <= value when minimizing, best >= value when maximizing). Returns False while the archive is empty.

Parameters:: value (float) – Target objective value.
Returns:: A composable condition that returns True when the best objective in the archive reaches value.
Return type:: TerminationCondition

Examples

>>> termination = Termination(max_fe(2000), f_target(1e-6))

saealib.stalled(window, tol=1e-08)[source]#

Create a termination condition based on lack of improvement (stagnation).

Terminates when the best score (f @ direction, higher is better) has not improved by more than tol for window consecutive generations. Improvement is tracked across calls using ctx.gen; the returned condition is therefore stateful and intended to be used once per run.

Parameters:

window (int) – Number of generations without improvement before terminating.
tol (float, optional) – Minimum score increase counted as an improvement, by default 1e-8.

Returns:

A composable condition that returns True after window stagnant generations.

Return type:

TerminationCondition

Examples

>>> termination = Termination(max_fe(2000), stalled(20))