def main():
args = parse_args()
if args.very_verbose:
args.verbose = args.very_verbose
warnings.simplefilter("default", textworld.TextworldGenerationWarning)
if args.seed is None:
args.seed = np.random.randint(65635)
grammar_flags = {
"theme": args.theme,
"include_adj": args.include_adj,
"only_last_action": args.only_last_action,
"blend_instructions": args.blend_instructions,
"blend_descriptions": args.blend_descriptions,
"ambiguous_instructions": args.ambiguous_instructions
}
print("Random seed: {}".format(args.seed))
rng = np.random.RandomState(args.seed)
agent = make_agent(args)
reward_history = []
for i in range(args.nb_games) if args.nb_games > 0 else itertools.count():
# Get a game seed to make everything reproducible.
game_seed = rng.randint(65635)
game_file, game = textworld.make(args.world_size,
args.nb_objects,
args.quest_length,
grammar_flags,
seed=game_seed,
games_dir=args.output)
print("Starting game {}".format(game_file))
env = textworld.start(game_file, backend=args.backend)
agent.reset(env)
if args.vizu >= 0:
from textworld.envs.wrappers import HtmlViewer
env = HtmlViewer(env, port=args.vizu)
game_state = env.reset()
if args.mode == "human" or args.verbose:
env.render()
reward = 0
done = False
for t in range(
args.max_steps) if args.max_steps > 0 else itertools.count():
command = agent.act(game_state, reward, done)
game_state, reward, done = env.step(command)
if args.mode == "human" or args.verbose:
env.render()
if done:
break
env.close()
print("Done after {} steps. Score {}/{}.".format(
game_state.nb_moves, game_state.score, game_state.max_score))
reward_history.append(reward)
if args.nb_games == 0: # Interactive mode.
input("Press enter to generate a new game.")
def main():
args = parse_args()
if args.very_verbose:
args.verbose = args.very_verbose
warnings.simplefilter("default", textworld.TextworldGenerationWarning)
if args.seed is None:
args.seed = np.random.randint(65635)
print("Random seed: {}".format(args.seed))
rng = np.random.RandomState(args.seed)
options = textworld.GameOptions()
options.grammar.theme = args.theme
options.grammar.include_adj = args.include_adj
options.grammar.only_last_action = args.only_last_action
options.grammar.blend_instructions = args.blend_instructions
options.grammar.blend_descriptions = args.blend_descriptions
options.grammar.ambiguous_instructions = args.ambiguous_instructions
options.nb_rooms = args.world_size
options.nb_objects = args.nb_objects
options.quest_length = args.quest_length
options.quest_breadth = args.quest_breadth
agent = make_agent(args)
reward_history = []
for i in range(args.nb_games) if args.nb_games > 0 else itertools.count():
options = options.copy()
options.seeds = rng.randint(65635)
game_file, game = textworld.make(options, args.output)
print("Starting game {}".format(game_file))
env = textworld.start(game_file)
agent.reset(env)
if args.vizu >= 0:
from textworld.envs.wrappers import HtmlViewer
env = HtmlViewer(env, port=args.vizu)
game_state = env.reset()
if args.mode == "human" or args.verbose:
env.render()
reward = 0
done = False
for t in range(
args.max_steps) if args.max_steps > 0 else itertools.count():
command = agent.act(game_state, reward, done)
game_state, reward, done = env.step(command)
if args.mode == "human" or args.verbose:
env.render()
if done:
break
env.close()
print("Done after {} steps. Score {}/{}.".format(
game_state.nb_moves, game_state.score, game_state.max_score))
reward_history.append(reward)
if args.nb_games == 0: # Interactive mode.
input("Press enter to generate a new game.")
def main():
args = build_parser().parse_args()
if args.very_verbose:
args.verbose = args.very_verbose
if args.state_fn:
state_wf = logging.StreamHandler(open(args.state_fn, "w"))
state_wf.setLevel(logging.INFO)
state_logger = logging.getLogger("states")
state_logger.setLevel(logging.INFO)
state_logger.addHandler(state_wf)
# open(args.state_fn, 'w')
if args.utts_fn:
# bc = logging.basicConfig(filename=args.utts_fn)
utts_wf = logging.StreamHandler(open(args.utts_fn, "w"))
# utts_wf = open(args.utts_fn, 'w')
else:
# bc = logging.basicConfig()
utts_wf = logging.StreamHandler(sys.stdout)
utts_wf.terminator = ''
utts_wf.setLevel(logging.INFO)
utts_logger = logging.getLogger("utterances")
utts_logger.setLevel(logging.INFO)
utts_logger.addHandler(utts_wf)
request_infos = textworld.EnvInfos(inventory=True,
description=True,
facts=True)
env = textworld.start(args.game, infos=request_infos)
if args.mode == "random":
agent = textworld.agents.NaiveAgent(seed=args.seed)
elif args.mode == "random-cmd":
# TODO get rid of redundancy
agent = textworld.agents.RandomCommandAgent(seed=args.seed)
elif args.mode == "human":
agent = textworld.agents.HumanAgent()
elif args.mode == 'walkthrough':
agent = textworld.agents.WalkthroughAgent()
agent.reset(env)
if args.viewer is not None:
from textworld.envs.wrappers import HtmlViewer
env = HtmlViewer(env, port=args.viewer)
if args.mode == "human" or args.very_verbose:
utts_logger.info("Using {}.\n".format(env.__class__.__name__))
game_state = env.reset()
if args.mode == "human" or args.verbose:
contents = env.render(mode="text")
contents = '\n'.join(
contents.split('\n')[22:]) # remove `TEXTWORLD` heading
contents = contents.replace('\n\n', '\n')
utts_logger.info(contents)
if args.state_fn:
state_logger.info(
json.dumps({
'facts':
[Proposition.serialize(prop) for prop in env.state.facts],
'inventory':
env.state.inventory.replace('\n\n', '\n'),
'description':
env.state.description.replace('\n\n', '\n'),
}))
reward = 0
done = False
for _ in range(
args.max_steps) if args.max_steps > 0 else itertools.count():
# tw_inform7.py
command = agent.act(game_state, reward, done)
# env.state.game.world.facts
game_state, reward, done = env.step(command)
if args.mode == "human" or args.verbose:
contents = env.render(mode="text")
contents = contents.replace('\n\n', '\n').replace('\n\n', '\n')
utts_logger.info(contents)
if args.state_fn:
state_logger.info(
json.dumps({
'facts': [
Proposition.serialize(prop)
for prop in env.state.facts
],
'inventory':
env.state.inventory.replace('\n\n', '\n'),
'description':
env.state.description.replace('\n\n', '\n'),
}))
if done:
break
env.close()
utts_logger.info("Done after {} steps. Score {}/{}.".format(
game_state.moves, game_state.score, game_state.max_score))