def get_achieving_streams(evaluations, stream_results, op=sum):
# TODO: could do this with bound_stream_instances instead
unprocessed_from_atom = defaultdict(list)
none = (None,) # None
node_from_atom = {none: Node(0, None)}
conditions_from_stream = {}
remaining_from_stream = {}
for stream_result in stream_results:
conditions_from_stream[stream_result] = stream_result.instance.get_domain() + (none,)
remaining_from_stream[stream_result] = len(conditions_from_stream[stream_result])
for atom in conditions_from_stream[stream_result]:
unprocessed_from_atom[atom].append(stream_result)
for atom in evaluations:
if is_atom(atom):
node_from_atom[fact_from_evaluation(atom)] = Node(0, None)
queue = [HeapElement(node.effort, atom) for atom, node in node_from_atom.items()]
while queue:
atom = heappop(queue).value
if atom not in unprocessed_from_atom:
continue
for stream_result in unprocessed_from_atom[atom]:
remaining_from_stream[stream_result] -= 1
if remaining_from_stream[stream_result]:
continue
effort = 1
total_effort = op(node_from_atom[cond].effort for cond in conditions_from_stream[stream_result]) + effort
for new_atom in stream_result.get_certified():
if (new_atom not in node_from_atom) or (total_effort < node_from_atom[new_atom].effort):
node_from_atom[new_atom] = Node(total_effort, stream_result)
heappush(queue, HeapElement(total_effort, new_atom))
del unprocessed_from_atom[atom]
del node_from_atom[none]
return node_from_atom
def proccess_stream_plan(key, sampling_problem, queue, evaluations, store):
instance, num_processed, bindings, stream_plan, action_plan, cost = sampling_problem
if not stream_plan:
store.add_plan(bind_plan(bindings, action_plan), cost)
return
if store.best_cost <= cost:
instance.disabled = False # TODO: only disable if not used elsewhere
# TODO: could just hash instances
return
opt_result = stream_plan[
0] # TODO: could do several at once but no real point
assert (not any(
evaluation_from_fact(f) not in evaluations
for f in instance.get_domain()))
# TODO: hash combinations to prevent repeats
results = []
for i in range(num_processed, len(instance.results_history)):
results.extend(instance.results_history[i])
if not results and not instance.enumerated:
#print(key.attempts, key.length)
results = instance.next_results(verbose=store.verbose)
for result in results:
add_certified(evaluations, result)
if (type(result) is PredicateResult) and (opt_result.value !=
result.value):
continue # TODO: check if satisfies target certified
new_bindings = bindings.copy()
if isinstance(result, StreamResult):
for opt, obj in zip(opt_result.output_objects,
result.output_objects):
assert (opt not in new_bindings
) # TODO: return failure if conflicting bindings
new_bindings[opt] = obj
new_stream_plan = stream_plan[1:]
new_cost = cost
if type(result) is FunctionResult:
new_cost += (result.value - opt_result.value)
new_key = SkeletonKey(0, len(new_stream_plan))
new_skeleton = Skeleton(
instantiate_first(new_bindings, new_stream_plan), 0, new_bindings,
new_stream_plan, action_plan, new_cost)
heappush(queue, HeapElement(new_key, new_skeleton))
if (key.attempts == 0) and isinstance(
opt_result, SynthStreamResult): # TODO: only add if failure?
new_stream_plan = opt_result.decompose() + stream_plan[1:]
new_key = SkeletonKey(0, len(new_stream_plan))
new_skeleton = Skeleton(instantiate_first(bindings,
new_stream_plan), 0,
bindings, new_stream_plan, action_plan, cost)
heappush(queue, HeapElement(new_key, new_skeleton))
if not instance.enumerated:
new_key = SkeletonKey(key.attempts + 1, len(
stream_plan)) # TODO: compute expected sampling effort required
new_skeleton = Skeleton(instance, len(instance.results_history),
bindings, stream_plan, action_plan, cost)
heappush(queue, HeapElement(new_key, new_skeleton))
def push_instance(self, instance):
# TODO: flush stale priorities?
complexity = self.compute_complexity(instance)
priority = Priority(complexity, self.num_pushes)
queue = self.function_queue if isinstance(
instance, FunctionInstance) else self.stream_queue
if isinstance(queue, deque):
queue.append(HeapElement(priority, instance))
else:
heappush(queue, HeapElement(priority, instance))
self.num_pushes += 1
def topological_sort(vertices, orders, priority_fn=lambda v: 0):
# Can also do a DFS version
incoming_edges, outgoing_edges = neighbors_from_orders(orders)
ordering = []
queue = []
for v in vertices:
if not incoming_edges[v]:
heappush(queue, HeapElement(priority_fn(v), v))
while queue:
v1 = heappop(queue).value
ordering.append(v1)
for v2 in outgoing_edges[v1]:
incoming_edges[v2].remove(v1)
if not incoming_edges[v2]:
heappush(queue, HeapElement(priority_fn(v2), v2))
return ordering
def push_instance(self, instance):
# TODO: flush stale priorities?
complexity = self.compute_complexity(instance)
priority = Priority(complexity, self.num_pushes)
heappush(self.queue, HeapElement(priority, instance))
self.num_pushes += 1
if self.verbose:
print(self.num_pushes, instance)
def add_skeleton(self, stream_plan, plan_attempts, bindings, plan_index,
cost):
stream_plan = [result.remap_inputs(bindings) for result in stream_plan]
attempted = sum(plan_attempts) != 0 # Bias towards unused
effort = compute_effort(plan_attempts)
key = SkeletonKey(attempted, effort)
skeleton = Skeleton(stream_plan, plan_attempts, bindings, plan_index,
cost)
heappush(self.queue, HeapElement(key, skeleton))
def get_achieving_streams(evaluations,
stream_results,
max_effort=INF,
**effort_args):
unprocessed_from_atom = defaultdict(list)
node_from_atom = {NULL_COND: Node(0, None)}
conditions_from_stream = {}
remaining_from_stream = {}
for result in stream_results:
conditions_from_stream[result] = result.instance.get_domain() + (
NULL_COND, )
remaining_from_stream[result] = len(conditions_from_stream[result])
for atom in conditions_from_stream[result]:
unprocessed_from_atom[atom].append(result)
for atom in evaluations:
if not is_negated_atom(atom):
node_from_atom[fact_from_evaluation(atom)] = Node(0, None)
queue = [
HeapElement(node.effort, atom)
for atom, node in node_from_atom.items()
]
while queue:
atom = heappop(queue).value
if atom not in unprocessed_from_atom:
continue
for result in unprocessed_from_atom[atom]:
remaining_from_stream[result] -= 1
if remaining_from_stream[result]:
continue
effort = result.get_effort(**effort_args)
total_effort = effort + EFFORT_OP(
node_from_atom[cond].effort
for cond in conditions_from_stream[result])
if max_effort <= total_effort:
continue
for new_atom in result.get_certified():
if (new_atom not in node_from_atom) or (
total_effort < node_from_atom[new_atom].effort):
node_from_atom[new_atom] = Node(total_effort, result)
heappush(queue, HeapElement(total_effort, new_atom))
del unprocessed_from_atom[atom]
del node_from_atom[NULL_COND]
return node_from_atom
def get_achieving_streams(evaluations,
stream_results,
unit_efforts=False): #, max_effort=INF):
# TODO: could do this with bound_stream_instances instead
unprocessed_from_atom = defaultdict(list)
node_from_atom = {NULL_COND: Node(0, None)}
conditions_from_stream = {}
remaining_from_stream = {}
for result in stream_results:
conditions_from_stream[result] = result.instance.get_domain() + (
NULL_COND, )
remaining_from_stream[result] = len(conditions_from_stream[result])
for atom in conditions_from_stream[result]:
unprocessed_from_atom[atom].append(result)
for atom in evaluations:
if is_atom(atom):
node_from_atom[fact_from_evaluation(atom)] = Node(0, None)
queue = [
HeapElement(node.effort, atom)
for atom, node in node_from_atom.items()
]
while queue:
atom = heappop(queue).value
if atom not in unprocessed_from_atom:
continue
for result in unprocessed_from_atom[atom]:
remaining_from_stream[result] -= 1
if remaining_from_stream[result]:
continue
effort = get_instance_effort(result.instance, unit_efforts)
total_effort = effort + COMBINE_OP(
node_from_atom[cond].effort
for cond in conditions_from_stream[result])
#if max_effort <= total_effort:
# continue
for new_atom in result.get_certified():
if (new_atom not in node_from_atom) or (
total_effort < node_from_atom[new_atom].effort):
node_from_atom[new_atom] = Node(total_effort, result)
heappush(queue, HeapElement(total_effort, new_atom))
del unprocessed_from_atom[atom]
del node_from_atom[NULL_COND]
return node_from_atom
def add_skeleton(self, stream_plan, plan_attempts, bindings, plan_index,
cost):
stream_plan = instantiate_plan(bindings, stream_plan)
#score = score_stream_plan(stream_plan)
attempted = sum(plan_attempts) != 0
effort = compute_effort(plan_attempts)
#effort = compute_score(plan_attempts)
#effort = compute_score2(plan_attempts)
key = SkeletonKey(attempted, effort)
skeleton = Skeleton(stream_plan, plan_attempts, bindings, plan_index,
cost)
heappush(self.queue, HeapElement(key, skeleton))
def get_element(self):
# TODO: instead of remaining, use the index in the queue to reprocess earlier ones
# TODO: include the complexity here as well
remaining = len(self.skeleton.stream_plan) - self.index
priority = Priority(self.attempts, remaining, self.cost)
return HeapElement(priority, self)
def get_element(self):
return HeapElement(self.get_priority(), self)
def solve_focused(problem,
stream_info={},
action_info={},
synthesizers=[],
max_time=INF,
max_cost=INF,
unit_costs=None,
sampling_time=0,
effort_weight=None,
eager_layers=1,
visualize=False,
verbose=True,
postprocess=False,
**search_kwargs):
"""
Solves a PDDLStream problem by first hypothesizing stream outputs and then determining whether they exist
:param problem: a PDDLStream problem
:param action_info: a dictionary from stream name to ActionInfo for planning and execution
:param stream_info: a dictionary from stream name to StreamInfo altering how individual streams are handled
:param max_time: the maximum amount of time to apply streams
:param max_cost: a strict upper bound on plan cost
:param effort_weight: a multiplier for stream effort compared to action costs
:param eager_layers: the number of eager stream application layers per iteration
:param visualize: if True, it draws the constraint network and stream plan as a graphviz file
:param verbose: if True, this prints the result of each stream application
:param search_kwargs: keyword args for the search subroutine
:return: a tuple (plan, cost, evaluations) where plan is a sequence of actions
(or None), cost is the cost of the plan, and evaluations is init but expanded
using stream applications
"""
# TODO: return to just using the highest level samplers at the start
solve_stream_plan_fn = relaxed_stream_plan if effort_weight is None else simultaneous_stream_plan
# TODO: warning check if using simultaneous_stream_plan or sequential_stream_plan with non-eager functions
num_iterations = 0
store = SolutionStore(max_time, max_cost,
verbose) # TODO: include other info here?
evaluations, goal_expression, domain, stream_name, externals = parse_problem(
problem, stream_info)
compile_to_exogenous(evaluations, domain, externals)
if unit_costs is None:
unit_costs = not has_costs(domain)
full_action_info = get_action_info(action_info)
load_stream_statistics(stream_name, externals + synthesizers)
if visualize:
clear_visualizations()
eager_externals = list(filter(lambda e: e.info.eager, externals))
streams, functions, negative = partition_externals(externals)
queue = []
# TODO: switch to searching if believe chance of search better than sampling
while not store.is_terminated():
num_iterations += 1
# TODO: decide max_sampling_time based on total search_time or likelihood estimates
print(
'\nIteration: {} | Queue: {} | Evaluations: {} | Cost: {} | Time: {:.3f}'
.format(num_iterations, len(queue), len(evaluations),
store.best_cost, store.elapsed_time()))
layered_process_stream_queue(
Instantiator(evaluations, eager_externals), evaluations, store,
eager_layers)
solve_stream_plan = lambda sr: solve_stream_plan_fn(
evaluations,
goal_expression,
domain,
sr,
negative,
max_cost=store.best_cost,
#max_cost=min(store.best_cost, max_cost),
unit_costs=unit_costs,
**search_kwargs)
#combined_plan, cost = solve_stream_plan(populate_results(evaluations, streams + functions))
combined_plan, cost = iterative_solve_stream_plan(
evaluations, streams, functions, solve_stream_plan)
if action_info:
combined_plan = reorder_combined_plan(evaluations, combined_plan,
full_action_info, domain)
print('Combined plan: {}'.format(combined_plan))
stream_plan, action_plan = separate_plan(combined_plan,
full_action_info)
stream_plan = reorder_stream_plan(
stream_plan) # TODO: is this strictly redundant?
stream_plan = get_synthetic_stream_plan(stream_plan, synthesizers)
print('Stream plan: {}\n'
'Action plan: {}'.format(stream_plan, action_plan))
if stream_plan is None:
if queue:
fairly_process_queue(queue, evaluations, store)
else:
break
else:
if visualize:
create_visualizations(evaluations, stream_plan, num_iterations)
heappush(
queue,
HeapElement(
SkeletonKey(0, len(stream_plan)),
Skeleton(instantiate_first({}, stream_plan), 0, {},
stream_plan, action_plan, cost)))
greedily_process_queue(queue, evaluations, store, sampling_time)
if postprocess and (not unit_costs):
locally_optimize(evaluations, store, goal_expression, domain,
functions, negative, synthesizers)
write_stream_statistics(stream_name, externals + synthesizers, verbose)
return revert_solution(store.best_plan, store.best_cost, evaluations)
def push_binding(self, binding):
# TODO: add to standby if not active
priority = binding.get_priority()
element = HeapElement(priority, binding)
heappush(self.queue, element)
