SPSA

class SPSA(max_trials=1000, save_steps=1, last_avg=1, c0=0.6283185307179586, c1=0.1, c2=0.602, c3=0.101, c4=0, skip_calibration=False)

Simultaneous Perturbation Stochastic Approximation (SPSA) optimizer.

SPSA is an algorithmic method for optimizing systems with multiple unknown parameters. As an optimization method, it is appropriately suited to large-scale population models, adaptive modeling, and simulation optimization.

See also

Many examples are presented at the SPSA Web site.

SPSA is a descent method capable of finding global minima, sharing this property with other methods as simulated annealing. Its main feature is the gradient approximation, which requires only two measurements of the objective function, regardless of the dimension of the optimization problem.

Note

SPSA can be used in the presence of noise, and it is therefore indicated in situations involving measurement uncertainty on a quantum computation when finding a minimum. If you are executing a variational algorithm using a Quantum ASseMbly Language (QASM) simulator or a real device, SPSA would be the most recommended choice among the optimizers provided here.

The optimization process includes a calibration phase, which requires additional functional evaluations.

For further details, please refer to https://arxiv.org/pdf/1704.05018v2.pdf#section*.11 (Supplementary information Section IV.)

Parameters

• max_trials (int) – Maximum number of iterations to perform.

• save_steps (int) – Save intermediate info every save_steps step. It has a min. value of 1.

• last_avg (int) – Averaged parameters over the last_avg iterations. If last_avg = 1, only the last iteration is considered. It has a min. value of 1.

• c0 (float) – The initial a. Step size to update parameters.

• c1 (float) – The initial c. The step size used to approximate gradient.

• c2 (float) – The alpha in the paper, and it is used to adjust a (c0) at each iteration.

• c3 (float) – The gamma in the paper, and it is used to adjust c (c1) at each iteration.

• c4 (float) – The parameter used to control a as well.

• skip_calibration (float) – Skip calibration and use provided c(s) as is.

Attributes

SPSA.bounds_support_level	Returns bounds support level
SPSA.gradient_support_level	Returns gradient support level
SPSA.initial_point_support_level	Returns initial point support level
SPSA.is_bounds_ignored	Returns is bounds ignored
SPSA.is_bounds_required	Returns is bounds required
SPSA.is_bounds_supported	Returns is bounds supported
SPSA.is_gradient_ignored	Returns is gradient ignored
SPSA.is_gradient_required	Returns is gradient required
SPSA.is_gradient_supported	Returns is gradient supported
SPSA.is_initial_point_ignored	Returns is initial point ignored
SPSA.is_initial_point_required	Returns is initial point required
SPSA.is_initial_point_supported	Returns is initial point supported
SPSA.setting	Return setting

Methods

SPSA.get_support_level()	return support level dictionary
SPSA.gradient_num_diff(x_center, f, epsilon)	We compute the gradient with the numeric differentiation in the parallel way, around the point x_center.
SPSA.optimize(num_vars, objective_function)	Perform optimization.
SPSA.print_options()	Print algorithm-specific options.
SPSA.set_max_evals_grouped(limit)	Set max evals grouped
SPSA.set_options(**kwargs)	Sets or updates values in the options dictionary.
SPSA.wrap_function(function, args)	Wrap the function to implicitly inject the args at the call of the function.