SequenceDesign

class blochsim.optim.SequenceDesign

Bases: Module

A cost, the parameters it is minimized over, and their limits.

Parameters:

• cost (callable) – Called with the designed parameters by keyword, each a tensor inside its limits; returns a one-element tensor, differentiable with respect to them.

• **parameters (Bounded or array-like) – One entry per designed parameter, under the keyword the cost reads it by. A Bounded carries its limits; a bare array is unbounded. At least one.

Raises:

ValueError – If no parameter is given, or a lower limit is not below its upper.

Notes

A preference (a soft penalty) belongs in the cost; a hard scanner limit belongs in Bounded.

Examples

design = SequenceDesign(
    sharpness, flip=Bounded(torch.full((8, 120), 120.0), 20.0, 180.0)
)
result = design.minimize(iterations=120)

Methods

minimize	Minimize the cost over the designed parameters.
values	Return the parameters as passed to the cost, inside their limits.

values()

Return the parameters as passed to the cost, inside their limits.

forward()

Evaluate the cost at the current parameters.

Returns:

The cost, one number, attached to the autograd graph.

Return type:

torch.Tensor

Raises:

ValueError – If the cost does not return one number.

minimize()

Minimize the cost over the designed parameters.

Parameters:

• iterations (int, default=200) – Number of updates, at least 1.

• learning_rate (float, default=0.05) – Adam step size on the unconstrained variables, used when

• 0. (optimizer_factory is None. Above)

• optimizer_factory (callable, default=None) – Maps the list of unconstrained variables to a torch.optim.Optimizer. None is Adam.

• callback (callable, default=None) – Called after each update as callback(step, parameters, loss) with detached values; returning True stops early.

Returns:

Parameters after the last update, inside their limits, and the cost before each update. Early stopping shortens the history.

Return type:

SequenceOptimization

Raises:

ValueError – If iterations or learning_rate is not positive.