def rollout(self, lookahead, env, n: OR_Node):
while not n.SOLVED and lookahead.sim_calls - lookahead.init_sim_calls < lookahead.sim_budget:
lookahead.rollout_depth += 1
t0 = time.perf_counter()
lookahead.expand(env, n)
# Pick random unsolved child of n
t0 = time.perf_counter()
n = lookahead.pick_random_unsolved_child(env, n)
tf = time.perf_counter()
lookahead.rollout_runtime_pick_random_unsolved += tf - t0
if n.terminal:
n.visited = True
lookahead.num_visited += 1
lookahead.solve_and_propagate_labels(n)
if lookahead.worst_terminal_accumulated_reward is None or lookahead.worst_terminal_accumulated_reward > n.accumulated_reward:
lookahead.worst_terminal_accumulated_reward = n.accumulated_reward
break
t0 = time.perf_counter()
is_novel = lookahead.root.feature_table.is_novel((n.state[0]))
tf = time.perf_counter()
lookahead.rollout_runtime_is_novel += tf - t0
if is_novel:
n.visited = True
lookahead.num_visited += 1
lookahead.root.feature_table.update_feature_table((n.state[0]))
elif not n.visited:
#pruned as is not novel
n.randomV = lookahead.cost_to_go_est(env, n)
lookahead.solve_and_propagate_labels(n)
break
if not n.SOLVED and lookahead._pruned_state_strategy == "heuristic":
# If didn't finish rollout due to computational budget apply heuistic value
n.randomV = lookahead.cost_to_go_est(env, n)
def rollout(self, lookahead, env, n: OR_Node):
while not n.SOLVED and lookahead.sim_calls - lookahead.init_sim_calls < lookahead.sim_budget:
lookahead.rollout_depth += 1
lookahead.expand(env, n)
# Pick random unsolved child of n
n = lookahead.pick_random_unsolved_child(env, n)
if n.terminal:
n.visited = True
lookahead.num_visited += 1
lookahead.solve_and_propagate_labels(n)
if lookahead.worst_terminal_accumulated_reward is None or lookahead.worst_terminal_accumulated_reward > n.accumulated_reward:
lookahead.worst_terminal_accumulated_reward = n.accumulated_reward
break
f, v, rank, old_depth = lookahead.root.feature_table.get_novel_feature(
(n.state[0], n.d))
if n.d < old_depth:
n.visited = True
lookahead.num_visited += 1
lookahead.root.feature_table.update_feature_table(
(n.state[0], n.d))
elif not n.visited and n.d >= old_depth:
n.visited = True
lookahead.num_visited += 1
#pruned as is not novel
n.randomV = lookahead.cost_to_go_est(env, n)
lookahead.solve_and_propagate_labels(n)
break
elif n.visited and old_depth < n.d:
#pruned as is not novel
n.randomV = lookahead.cost_to_go_est(env, n)
n._children = {}
lookahead.solve_and_propagate_labels(n)
break
if not n.SOLVED and lookahead._pruned_state_strategy == "heuristic": #If didn't finish rollout due to computational budget apply heuistic value
n.randomV = lookahead.cost_to_go_est(env, n)