Class solver::PopulationAnnealing

Population Annealing.

Population based optimizer with variable beta stepping. PopulationAnnealing explores the configuration space using multiple Metropolis walkers subject to a birth-death process (depending on their relative cost-fitness within the population). Stepping method Population annealing supports three different resampling_strategys for dynamically selecting the next step in temperature space (delta_beta):

friction_tensor: The stepping size is calculated from

 \Delta\beta = c_{FT} / \sqrt\zeta

{
  'target': 'populationannealing.cpu',
  'version': '1.0',
  'resampling_strategy': 'friction_tensor',
  'friction_tensor_constant': 1
}

energy_variance: Scale temperature stepping with the variance in the energy (cost):

 \Delta\beta = f_{culling} / \sqrt\sigma

for this, the initial value for the culling fraction needs to be specified:

{
  'target': 'populationannealing.cpu',
  'version': '1.0',
  'resampling_strategy': 'energy_variance',
  'initial_culling_fraction': 0.5
}

constant_culling: The stepping size is chosen to keep the culling rate constant at culling_fraction (by estimating the expected death ratio and solving for delta_beta):

{
  'target': 'populationannealing.cpu',
  'version': '1.0',
  'resampling_strategy': 'constant_culling',
  'culling_fraction': 0.1
}

Inheritance

solver::SteppingSolver

solver::PopulationAnnealing

solver::PAParameterFree

Inherited Members

clear_observable_label

restart

is_watching

scoped_observable_label

observe

Observer

render

set_observable_label

~ModelSolver

get_solutions

copy_lowest_state

copy_solutions

get_model_properties

update_lowest_cost

get_model_term_size

init_memory_check

count_solutions

copy_solutions

get_model

get_model_unconst

copy_solutions_other

ModelSolver

populate_solutions

max_replicas_adjusted_state

set_model

get_model_sweep_size

is_empty

get_lowest_cost

adjust_states

max_replicas_of_state

set_time_limit

get_result

copy_limits

~Solver

get_benchmark

get_max_threads

get_thread_count

Solver

fixed_step_per_tick

update_accumulated_info

handle_signals

run

current_steps

reset

get_cost_function_evaluation_count

SteppingSolver

get_solver_properties

status

get_seed

~SteppingSolver

get_runtime

seconds_per_step

set_output_parameter

param

get_output_parameters

~Component

Component

get_status

get_class_name

Constructors

PopulationAnnealing()

Declaration

solver::PopulationAnnealing<Model_T>::PopulationAnnealing()

PopulationAnnealing()

Declaration

solver::PopulationAnnealing<Model_T>::PopulationAnnealing(const PopulationAnnealing&)=delete

Methods

operator=()

Declaration

PopulationAnnealing&solver::PopulationAnnealing<Model_T>::operator=(const PopulationAnnealing&)=delete

get_identifier()

Identifier of this solver (target in the request)

Declaration

std::string solver::PopulationAnnealing<Model_T>::get_identifier() const override

init_memory_check_error_message()

Declaration

std::string solver::PopulationAnnealing<Model_T>::init_memory_check_error_message() const override

target_number_of_states()

Declaration

size_t solver::PopulationAnnealing<Model_T>::target_number_of_states() const override

init()

Initialize the solver.

Declaration

void solver::PopulationAnnealing<Model_T>::init() override

init_population()

Reset the population with size target_population

Note

This method is invoked both at the start of the simulation and whenever a restart/population increase is triggered.

Declaration

void solver::PopulationAnnealing<Model_T>::init_population()

find_delta_beta()

Adjust the stepping size delta_beta until the estimator returns an estimate of zero. NOTE: This is NOT a generic root finder, it makes several assumptions about the estimator being used.

Declaration

double solver::PopulationAnnealing<Model_T>::find_delta_beta(T&estimator)

make_step()

Declaration

void solver::PopulationAnnealing<Model_T>::make_step(uint64_t step) override

resample()

Declaration

void solver::PopulationAnnealing<Model_T>::resample(double delta_beta)

configure()

Check the identifier and version against the configuraiton.

Declaration

void solver::PopulationAnnealing<Model_T>::configure(const utils::Json&json) override

finalize()

Declaration

void solver::PopulationAnnealing<Model_T>::finalize() override

expand_population()

expand the population to size target_population

Note

This method is called to expand population from current size to target_population

Declaration

void solver::PopulationAnnealing<Model_T>::expand_population()

is_linear_schedule()

Declaration

bool solver::PopulationAnnealing<Model_T>::is_linear_schedule() const

is_geometric_schedule()

Declaration

bool solver::PopulationAnnealing<Model_T>::is_geometric_schedule() const

is_friction_tensor()

Declaration

bool solver::PopulationAnnealing<Model_T>::is_friction_tensor() const

is_energy_variance()

Declaration

bool solver::PopulationAnnealing<Model_T>::is_energy_variance() const

is_constant_culling()

Declaration

bool solver::PopulationAnnealing<Model_T>::is_constant_culling() const