saealib.InequalityConstraint¶

class saealib.InequalityConstraint(func, threshold=0.0)[source]¶

Bases: object

A single inequality constraint: g(x) <= threshold.

Parameters:

func (callable) – Constraint function g(x). Should return a scalar float.
threshold (float, optional) – Right-hand side of the constraint g(x) <= threshold. Default: 0.0.

Methods

`__init__`
`evaluate`	Return the raw constraint value g(x).
`evaluate_with_violation`	Return the raw value and violation with a single function evaluation.
`gradient`	Return the gradient of g(x) with respect to x, if available.
`violation`	Return constraint violation: max(0, g(x) - threshold).
`violation_from_value`	Return the violation for an already-evaluated raw value g(x).

Method Details

InequalityConstraint.__init__(func, threshold=0.0)[source]¶

Parameters:

func (callable)
threshold (float)

InequalityConstraint.evaluate(x)[source]¶

Return the raw constraint value g(x).

Parameters:: x (ndarray)
Return type:: float

InequalityConstraint.evaluate_with_violation(x)[source]¶

Return the raw value and violation with a single function evaluation.

Returns:

g (float) – Raw constraint value g(x).
cv (float) – Constraint violation: max(0, g(x) - threshold).

Parameters:

x (ndarray)

Return type:

tuple[float, float]

InequalityConstraint.gradient(x)[source]¶

Return the gradient of g(x) with respect to x, if available.

Override this in a subclass to enable gradient-based constraint handlers (e.g. gradient-based repair). The default returns None, signalling that no analytical Jacobian is provided.

Parameters:: x (np.ndarray) – The solution at which to evaluate the gradient. shape = (dim, )
Returns:: Gradient vector with shape = (dim, ), or None when no gradient is available.
Return type:: np.ndarray or None

InequalityConstraint.violation(x)[source]¶

Return constraint violation: max(0, g(x) - threshold).

Parameters:: x (ndarray)
Return type:: float

InequalityConstraint.violation_from_value(g)[source]¶

Return the violation for an already-evaluated raw value g(x).

This is the single source of truth for the per-constraint violation formula: violation(), evaluate_with_violation(), and StaticToleranceHandler all delegate to it. Subclasses (e.g. EqualityConstraint) override this one method to change how a raw value maps to a violation, without re-evaluating func.

Parameters:: g (float) – Raw constraint value g(x).
Returns:: Constraint violation: max(0, g - threshold).
Return type:: float