kwargs.jl

OptimOptions

Keyword arguments for optimise_design, and specifically the optimisation functions within it, including for the gradient descent function optimise_weights, the repetition number optimisation function optimise_repetitions, and the tuple set optimisation function optimise_tuple_set.

General options

• ls_type::Symbol = :gls: Type of least squares estimator for which we optimise the design, which can be :gls, :wls, or :ols.
• est_type::Symbol = :prod: Type of estimator for which we optimise the design, which can be :ordinary, :marginal, :relative, or :sum or :prod, which optimise for the arithmetic or geometric mean, respectively, of the :ordinary and :relative estimator figures of merit weighted by est_weight.
• est_weight::Real = 0.5: Weighting of the :ordinary estimator figure of merit in the arithmetic or geometric mean when est_type is :sum or :prod, respectively, such that :relative is weighted by 1 - est_weight.
• save_data::Bool = false: Whether to automatically save the optimised design.

Gradient descent options

• learning_rate::Real = (ls_type == :ols ? 1 : 10.0^(3/4)): Learning rate for the gradient descent algorithm.
• momentum::Real = 0.99: Momentum for the gradient descent algorithm.
• learning_rate_scale_factor::Real = 10.0^(1/4): Factor by which to reduce the learning rate if the gradient descent algorithm repeatedly steps in directions that reduce the figure of merit.
• max_steps::Integer = 1000: Maximum number of gradient descent steps to take.
• convergence_threshold::Real = 1e-5: Convergence threshold for the gradient descent algorithm.
• convergence_steps::Integer = 5: Number of steps over which to check convergence.
• grad_diagnostics::Bool = false: Whether to display gradient descent diagnostics.

Reptition number optimisation options

• rep_est_type::Symbol = :relative: Type of estimator for which we optimise the repetition number, which can be :ordinary, :marginal, :relative, or :sum or :prod, which optimise for the arithmetic or geometric mean, respectively, of the :ordinary and :relative estimator figures of merit weighted by rep_est_weight.
• rep_est_weight::Real = 0.5: Weighting of the :ordinary estimator figure of merit in the arithmetic or geometric mean when rep_est_type is :sum or :prod, respectively, such that :relative is weighted by 1 - rep_est_weight.
• error_target::Real = 0.1: Controls initialisation of the repetition numbers, which heuristically roughly target this error rate; note that setting this to 0 initialises repetition numbers as 0, and all such repetition numbers are then set to 2 before optimisation.
• add_circuit::Bool = true: Whether to add the circuit itself to the repeat tuple set.
• cycle_convergence_threshold::Real = 1e-5: Convergence threshold for the cyclic coordinate descent algorithm for optimising repetition numbers.
• convergence_cycles::Integer = 5: Number of steps over which to check convergence for the cyclic coordinate descent algorithm for optimising repetition numbers.
• min_depth::Integer = 0: Minimum depth of repeated tuples.
• max_depth::Integer = 1000: Maximum depth of repeated tuples.
• max_cycles::Integer = 100: Maximum number of cycles to use in the cyclic coordinate descent algorithm for optimising repetition numbers.
• repeat_points::Integer = 1: Each repeat number is augmented to have repeat_points repeat numbers, logarithmically spaced with the smallest repeat number which is shrunk by a factor initial_shrink_factor.
• initial_shrink_factor::Real = 2^(repeat_points - 1): Factor by which the smallest repeat number in an augmented tuple set is shrunk.
• rep_diagnostics::Bool = true: Whether to display repetition number optimisation diagnostics.

Tuple set optimisation options

• tuple_est_type::Symbol = :ordinary: Type of estimator for which we optimise the tuple set, which can be :ordinary, :marginal, :relative, :sum, or :prod, which optimise for the arithmetic or geometric mean, respectively, of the :ordinary and :relative estimator figures of merit weighted by tuple_est_weight.
• tuple_est_weight::Real = 0.5: Weighting of the :ordinary estimator figure of merit in the arithmetic or geometric mean when tuple_est_type is :sum or :prod, respectively, such that :relative is weighted by 1 - tuple_est_weight.
• max_excursions::Integer = 100: Number of excurisons used to optimise the tuple set.
• excursions_unchanged::Integer = 3: Number of excursions that must not change the tuple set before the optimisation routine terminates.
• excursion_length::Integer = 5: Number of tuples added by each excursion.
• extra_tuple_number::Integer = 5: Number of tuples beyond the number in the basic tuple set added by the tuple set optimisation procedure.
• max_tuple_length::Integer = 20: Maximum length of random tuples.
• tuple_length_zipf_power::Real = 1: Zipf power to use when Zipf-randomly choosing the length of random tuples.
• repeat_zipf_powers::Vector{Float64} = [Inf; 2.0]: Zipf power to use, chosen uniformly at random from the vector, when Zipf-randomly choosing how many times to repeat entries that will be appended to the end of a random tuple during its generation.
• mirror_values::Vector{Bool} = [false; true]: Whether to mirror the tuple, chosen uniformly at random from the vector, when generating random tuples.
• trial_factor::Integer = 20: Number of random tuples trialled for each tuple the excursion needs to add to the tuple set to grow it.
• grow_greedy::Bool = true: Whether the excursions add tuples to the set greedily according to the figure of merit, or to add them even if this reduces the figure of merit.
• weight_experiments::Bool = false: Whether to weight the shot weights for a tuple by the number of experiments for that tuple; while usually true, false performs better for tuple set optimisation.
• seed::Union{UInt64, Nothing} = nothing: Seed used to randomly generate tuples.
• tuple_diagnostics::Bool = true: Whether to display tuple set optimisation diagnostics.