def rec(state, depth, stack_size):
if clock() - start > time_limit:
return
s = state.get_score_abort()
if depth <= 0 or stack_size <= 0:
check(s)
return
preprocessed = preprocess_world(state)
if upper_bound(preprocessed)-state.time <= best.score:
return
aggressive_preprocess(preprocessed)
frozen_state = preprocessed.get_hash()
old_score = visited.get(frozen_state)
if old_score is not None and s <= old_score:
return
visited[frozen_state] = s
check(s)
zzz = 'LRUDW'
num_commands = len(commands)
next_steps = set()
greedy = path_to_nearest_lambda_or_lift(state)
if greedy is not None:
greedy = [greedy[1]]
else:
greedy = []
#greedy = []
for cmds in chain(greedy, product(zzz, zzz)):
new_state = state
for cmd in cmds:
if new_state.final_score is None:
new_state = new_state.apply_command(cmd)
commands.append(cmd)
if new_state.final_score is None:
h = new_state.get_hash()
if h not in next_steps:
next_steps.add(h)
if cmds in greedy:
new_depth = depth
else:
new_depth = depth-1
rec(new_state, new_depth, stack_size-1)
else:
check(new_state.final_score)
for _ in cmds:
commands.pop()
assert num_commands == len(commands)
def get_ub_and_cache_entry(self, world):
'''
return upper bound and CacheEntry;
given world is raw (not even preprocessed);
calls are cached
'''
assert not world.terminated
raw_hash = world.get_hash()
result = self.ub_ce_cache.get(raw_hash)
if result is not None:
self.stats['ub ce hit'] += 1
ub, cache_entry = result
return ub-world.time, cache_entry
self.check_stop()
preprocessed = preprocess_world(world)
self.stats['preprocess'] += 1
ub = upper_bound(preprocessed)
aggressive_preprocess(preprocessed) #inplace
key = preprocessed.get_hash()
cache_entry = self.cache.get(key)
if cache_entry is None:
cache_entry = self.cache[key] = CacheEntry()
cache_entry.time = 100000000
cache_entry.command = None
self.ub_ce_cache[raw_hash] = ub, cache_entry
return ub-world.time, cache_entry
