def simulated_annealing(constraint_problem: ConstraintProblem, max_steps: int, temperature: float, cooling_rate: float,
generate_start_state: StartStateGenerator = generate_start_state_randomly,
generate_successor: SuccessorGenerator = alter_random_variable_value_pair,
calculate_score: ScoreCalculator = consistent_constraints_amount) -> ConstraintProblem:
generate_start_state(constraint_problem)
if constraint_problem.is_completely_consistently_assigned():
return constraint_problem
max_steps -= 1
best_score = calculate_score(constraint_problem)
best_score_problem = deepcopy(constraint_problem)
for i in range(max_steps):
if constraint_problem.is_completely_consistently_assigned():
return constraint_problem
curr_score = calculate_score(constraint_problem)
if best_score < curr_score:
best_score = curr_score
best_score_problem = deepcopy(constraint_problem)
successor = generate_successor(constraint_problem)
successor_score = calculate_score(successor)
delta = successor_score - curr_score
if delta > 0 or uniform(0, 1) < exp(delta / temperature):
constraint_problem = successor
temperature *= cooling_rate
return best_score_problem
def random_restart_first_choice_hill_climbing(
constraint_problem: ConstraintProblem,
max_restarts: int,
max_steps: int,
max_successors: int,
generate_start_state: StartStateGenerator = generate_start_state_randomly,
generate_successor: SuccessorGenerator = alter_random_variable_value_pair,
calculate_score: ScoreCalculator = consistent_constraints_amount
) -> ConstraintProblem:
generate_start_state(constraint_problem)
if constraint_problem.is_completely_consistently_assigned():
return constraint_problem
max_restarts -= 1
best_score = calculate_score(constraint_problem)
best_score_problem = deepcopy(constraint_problem)
for i in range(max_restarts):
generate_start_state(constraint_problem)
for j in range(max_steps):
if constraint_problem.is_completely_consistently_assigned():
return constraint_problem
current_score = calculate_score(constraint_problem)
if best_score < current_score:
best_score = current_score
best_score_problem = deepcopy(constraint_problem)
for k in range(max_successors):
successor = generate_successor(constraint_problem)
successor_score = calculate_score(successor)
if current_score < successor_score:
constraint_problem = successor
break
return best_score_problem
def min_conflicts(
constraint_problem: ConstraintProblem,
max_steps: int,
tabu_size: int = -1,
with_history: bool = False) -> Optional[Deque[Tuple[Variable, Any]]]:
__tabu_queue.clear()
read_only_variables = constraint_problem.get_assigned_variables()
if tabu_size == -1:
tabu_size = 0
assert tabu_size + len(read_only_variables) < len(constraint_problem.get_variables()), \
"tabu_size + len(read_only_variables) is equal or bigger than constraint_problem's variables amount."
if tabu_size == 0:
tabu_size = -1
actions_history = None
if with_history:
actions_history = deque()
rand_assignmt_history = constraint_problem.assign_variables_with_random_values(
read_only_variables, actions_history)
if with_history:
actions_history.extend(rand_assignmt_history)
best_min_conflicts = len(constraint_problem.get_unsatisfied_constraints())
best_min_conflicts_assignment = constraint_problem.get_current_assignment()
for i in range(max_steps):
if constraint_problem.is_completely_consistently_assigned():
return actions_history
conflicted_variable = __get_random_conflicted_variable(
constraint_problem, read_only_variables, tabu_size)
conflicted_variable.unassign()
if with_history:
actions_history.append((conflicted_variable, None))
min_conflicts_value = __get_min_conflicts_value(
constraint_problem, conflicted_variable)
conflicted_variable.assign(min_conflicts_value)
if with_history:
actions_history.append((conflicted_variable, min_conflicts_value))
if len(__tabu_queue) == tabu_size:
__tabu_queue.popleft()
if __tabu_queue:
__tabu_queue.append(conflicted_variable)
curr_conflicts_count = len(
constraint_problem.get_unsatisfied_constraints())
if curr_conflicts_count < best_min_conflicts:
best_min_conflicts = curr_conflicts_count
best_min_conflicts_assignment = constraint_problem.get_current_assignment(
)
constraint_problem.unassign_all_variables()
constraint_problem.assign_variables_from_assignment(
best_min_conflicts_assignment)
return actions_history
def naive_cycle_cutset(constraint_problem: ConstraintProblem, with_history: bool = False) \
-> Optional[Deque[Tuple[Variable, Any]]]:
actions_history = None
if with_history:
actions_history = deque()
variables = constraint_problem.get_variables()
read_only_variables = constraint_problem.get_assigned_variables()
constraints = list(constraint_problem.get_constraints())
constraints.sort(key=lambda constraint: len(constraint.variables), reverse=True)
constraint_graph = constraint_problem.get_constraint_graph_as_adjacency_list()
for i in range(1, len(constraints)):
cutset_constraints = constraints[:i]
cutset_variables = set()
for cutset_const in cutset_constraints:
cutset_variables.update(cutset_const.variables)
reduced_graph = {var: neighbors for var, neighbors in constraint_graph.items() if var not in cutset_variables}
for var in reduced_graph:
reduced_graph[var] -= cutset_variables
if __is_tree(reduced_graph):
consistent_assignments_list = __get_consistent_assignments(cutset_variables, cutset_constraints,
read_only_variables)
non_cutset_variables = variables - cutset_variables
non_cutset_vars_to_original_domains_map = {var: var.domain for var in non_cutset_variables}
for consist_assignment in consistent_assignments_list:
for var, value in zip(cutset_variables, consist_assignment):
if var not in read_only_variables:
var.assign(value)
if with_history:
actions_history.append((var, value))
for non_cutset_var in non_cutset_variables:
if non_cutset_var not in read_only_variables:
non_cutset_var.domain = list(constraint_problem.get_consistent_domain(non_cutset_var))
tree_csp_action_history = tree_csp_solver(constraint_problem, with_history)
if with_history:
actions_history.extend(tree_csp_action_history)
if constraint_problem.is_completely_consistently_assigned():
return actions_history
for var in variables:
if var not in read_only_variables:
var.unassign()
if with_history:
actions_history.append((var, None))
for var in non_cutset_vars_to_original_domains_map:
if var not in read_only_variables:
var.domain = non_cutset_vars_to_original_domains_map[var]
return actions_history
def constraints_weighting(constraint_problem: ConstraintProblem, max_tries: int, with_history: bool = False) \
-> Optional[Deque[Tuple[Variable, Any]]]:
actions_history = None
if with_history:
actions_history = deque()
constraints_weights = {
constraint: 1
for constraint in constraint_problem.get_constraints()
}
read_only_variables = constraint_problem.get_assigned_variables()
for i in range(max_tries):
constraint_problem.assign_variables_with_random_values(
read_only_variables)
last_reduction = float("inf")
while 0 < last_reduction:
if constraint_problem.is_completely_consistently_assigned():
return actions_history
reduction, variable, value = __get_best_reduction_variable_value(
constraint_problem, constraints_weights, read_only_variables)
variable.unassign()
if with_history:
actions_history.append((variable, None))
variable.assign(value)
if with_history:
actions_history.append((variable, value))
last_reduction = reduction
for unsatisfied_constraint in constraint_problem.get_unsatisfied_constraints(
):
constraints_weights[unsatisfied_constraint] += 1
if i != max_tries - 1:
constraint_problem.unassign_all_variables(read_only_variables)
return actions_history