def main():
parser = argparse.ArgumentParser()
parser.add_argument('filename',
type=str) # defaultIS.h5/snapshots/iter0000480
parser.add_argument('--vid', type=str, default='/tmp/madrl.mp4')
parser.add_argument('--deterministic', action='store_true', default=False)
parser.add_argument('--heuristic', action='store_true', default=False)
parser.add_argument('--evaluate', action='store_true', default=False)
parser.add_argument('--n_trajs', type=int, default=20)
parser.add_argument('--n_steps', type=int, default=500)
parser.add_argument('--same_con_pol', action='store_true')
args = parser.parse_args()
fh = FileHandler(args.filename)
env_map = TwoDMaps.rectangle_map(
*map(int, fh.train_args['map_size'].split(',')))
map_pool = [env_map]
# map_pool = np.load(
# os.path.join('/scratch/megorov/deeprl/MADRL/runners/maps/', os.path.basename(fh.train_args[
# 'map_file'])))
env = PursuitEvade(map_pool,
n_evaders=fh.train_args['n_evaders'],
n_pursuers=fh.train_args['n_pursuers'],
obs_range=fh.train_args['obs_range'],
n_catch=fh.train_args['n_catch'],
urgency_reward=fh.train_args['urgency'],
surround=bool(fh.train_args['surround']),
sample_maps=bool(fh.train_args['sample_maps']),
flatten=bool(fh.train_args['flatten']),
reward_mech='global',
catchr=fh.train_args['catchr'],
term_pursuit=fh.train_args['term_pursuit'])
if fh.train_args['buffer_size'] > 1:
env = ObservationBuffer(env, fh.train_args['buffer_size'])
hpolicy = None
if args.heuristic:
from heuristics.pursuit import PursuitHeuristicPolicy
hpolicy = PursuitHeuristicPolicy(env.agents[0].observation_space,
env.agents[0].action_space)
if args.evaluate:
minion = Evaluator(env, fh.train_args, args.n_steps, args.n_trajs,
args.deterministic,
'heuristic' if args.heuristic else fh.mode)
evr = minion(fh.filename,
file_key=fh.file_key,
same_con_pol=args.same_con_pol,
hpolicy=hpolicy)
from tabulate import tabulate
print(evr)
print(tabulate(evr, headers='keys'))
else:
minion = Visualizer(env, fh.train_args, args.n_steps, args.n_trajs,
args.deterministic, fh.mode)
rew, info = minion(fh.filename, file_key=fh.file_key, vid=args.vid)
pprint.pprint(rew)
pprint.pprint(info)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('filename',
type=str) # defaultIS.h5/snapshots/iter0000480
parser.add_argument('--vid', type=str, default='/tmp/madrl.mp4')
parser.add_argument('--deterministic', action='store_true', default=False)
parser.add_argument('--heuristic', action='store_true', default=False)
parser.add_argument('--evaluate', action='store_true', default=False)
parser.add_argument('--n_trajs', type=int, default=10)
parser.add_argument('--n_steps', type=int, default=500)
parser.add_argument('--same_con_pol', action='store_true')
args = parser.parse_args()
fh = FileHandler(args.filename)
env = MAWaterWorld(
fh.train_args['n_pursuers'],
fh.train_args['n_evaders'],
fh.train_args['n_coop'],
fh.train_args['n_poison'],
n_sensors=fh.train_args['n_sensors'],
food_reward=fh.train_args['food_reward'],
poison_reward=fh.train_args['poison_reward'],
reward_mech='global',
encounter_reward=0, #fh.train_args['encounter_reward'],
addid=True,
)
if fh.train_args['buffer_size'] > 1:
env = ObservationBuffer(env, fh.train_args['buffer_size'])
hpolicy = None
if args.heuristic:
from heuristics.waterworld import WaterworldHeuristicPolicy
hpolicy = WaterworldHeuristicPolicy(env.agents[0].observation_space,
env.agents[0].action_space)
if args.evaluate:
minion = Evaluator(env, fh.train_args, args.n_steps, args.n_trajs,
args.deterministic,
'heuristic' if args.heuristic else fh.mode)
evr = minion(fh.filename,
file_key=fh.file_key,
same_con_pol=args.same_con_pol,
hpolicy=hpolicy)
from tabulate import tabulate
print(tabulate(evr, headers='keys'))
else:
minion = Visualizer(env, fh.train_args, args.n_steps, args.n_trajs,
args.deterministic,
'heuristic' if args.heuristic else fh.mode)
rew, info = minion(fh.filename,
file_key=fh.file_key,
vid=args.vid,
hpolicy=hpolicy)
pprint.pprint(rew)
pprint.pprint(info)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('filename', type=str)
parser.add_argument('--vid', type=str, default='madrl.mp4')
parser.add_argument('--deterministic', action='store_true', default=False)
parser.add_argument('--heuristic', action='store_true', default=False)
parser.add_argument('--evaluate', action='store_true', default=True)
parser.add_argument('--n_trajs', type=int, default=20)
parser.add_argument('--n_steps', type=int, default=20)
parser.add_argument('--same_con_pol', action='store_true')
args = parser.parse_args()
fh = FileHandler(args.filename)
if fh.train_args['map_file'] is not None:
map_pool = np.load(
os.path.join('', os.path.basename(fh.train_args[
'map_file'])))
else:
if fh.train_args['map_type'] == 'rectangle':
env_map = TwoDMaps.rectangle_map(*map(int, fh.train_args['map_size'].split(',')))
elif args.map_type == 'complex':
env_map = TwoDMaps.complex_map(*map(int, fh.train_args['map_size'].split(',')))
else:
raise NotImplementedError()
map_pool = [env_map]
env = PursuitEvade(map_pool, n_evaders=fh.train_args['n_evaders'],
n_pursuers=fh.train_args['n_pursuers'], obs_range=fh.train_args['obs_range'],
n_catch=fh.train_args['n_catch'], urgency_reward=fh.train_args['urgency'],
surround=bool(fh.train_args['surround']),
sample_maps=bool(fh.train_args['sample_maps']),
flatten=bool(fh.train_args['flatten']), reward_mech='global',
catchr=fh.train_args['catchr'], term_pursuit=fh.train_args['term_pursuit'])
if fh.train_args['buffer_size'] > 1:
env = ObservationBuffer(env, fh.train_args['buffer_size'])
hpolicy = None
if args.evaluate:
minion = Evaluator(env, fh.train_args, args.n_steps, args.n_trajs, args.deterministic,
'heuristic' if args.heuristic else fh.mode)
evr = minion(fh.filename, file_key=fh.file_key, same_con_pol=args.same_con_pol,
hpolicy=hpolicy)
print(evr)
print(tabulate(evr, headers='keys'))
else:
minion = Visualizer(env, fh.train_args, args.n_steps, args.n_trajs, args.deterministic,
fh.mode)
rew, info = minion(fh.filename, file_key=fh.file_key, vid=args.vid)
pprint.pprint(rew)
pprint.pprint(info)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('filename',
type=str) # defaultIS.h5/snapshots/iter0000480
parser.add_argument('--vid', type=str, default='/tmp/madrl.mp4')
parser.add_argument('--deterministic', action='store_true', default=False)
parser.add_argument('--heuristic', action='store_true', default=False)
parser.add_argument('--evaluate', action='store_true', default=False)
parser.add_argument('--save_file', type=str, default=None)
parser.add_argument('--n_trajs', type=int, default=20)
parser.add_argument('--n_steps', type=int, default=500)
parser.add_argument('--same_con_pol', action='store_true')
args = parser.parse_args()
fh = FileHandler(args.filename)
env = MultiAnt(n_legs=fh.train_args['n_legs'],
ts=fh.train_args['ts'],
integrator=fh.train_args['integrator'],
out_file=fh.train_args['out_file'],
base_file=fh.train_args['base_file'],
reward_mech=fh.train_args['reward_mech'])
if fh.train_args['buffer_size'] > 1:
env = ObservationBuffer(env, fh.train_args['buffer_size'])
hpolicy = None
if args.evaluate:
minion = Evaluator(env, fh.train_args, args.n_steps, args.n_trajs,
args.deterministic,
'heuristic' if args.heuristic else fh.mode)
evr = minion(fh.filename,
file_key=fh.file_key,
same_con_pol=args.same_con_pol,
hpolicy=hpolicy)
if args.save_file:
pickle.dump(evr, open(args.save_file, "wb"))
from tabulate import tabulate
#print(tabulate(evr, headers='keys'))
else:
minion = Visualizer(env, fh.train_args, args.n_steps, args.n_trajs,
args.deterministic, fh.mode)
rew, info = minion(fh.filename, file_key=fh.file_key, vid=args.vid)
pprint.pprint(rew)
pprint.pprint(info)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('filename',
type=str) # defaultIS.h5/snapshots/iter0000480
parser.add_argument('--vid', type=str, default='/tmp/madrl.mp4')
parser.add_argument('--deterministic', action='store_true', default=False)
parser.add_argument('--heuristic', action='store_true', default=False)
parser.add_argument('--evaluate', action='store_true', default=False)
parser.add_argument('--n_trajs', type=int, default=20)
parser.add_argument('--n_steps', type=int, default=500)
parser.add_argument('--same_con_pol', action='store_true')
args = parser.parse_args()
fh = FileHandler(args.filename)
env = MultiWalkerEnv(fh.train_args['n_walkers'],
fh.train_args['position_noise'],
fh.train_args['angle_noise'],
reward_mech='global') #fh.train_args['reward_mech'])
if fh.train_args['buffer_size'] > 1:
env = ObservationBuffer(env, fh.train_args['buffer_size'])
hpolicy = None
if args.heuristic:
from heuristics.multiwalker import MultiwalkerHeuristicPolicy
hpolicy = MultiwalkerHeuristicPolicy(env.agents[0].observation_space,
env.agents[0].action_space)
if args.evaluate:
minion = Evaluator(env, fh.train_args, args.n_steps, args.n_trajs,
args.deterministic,
'heuristic' if args.heuristic else fh.mode)
evr = minion(fh.filename,
file_key=fh.file_key,
same_con_pol=args.same_con_pol,
hpolicy=hpolicy)
from tabulate import tabulate
print(tabulate(evr, headers='keys'))
else:
minion = Visualizer(env, fh.train_args, args.n_steps, args.n_trajs,
args.deterministic, fh.mode)
rew, info = minion(fh.filename, file_key=fh.file_key, vid=args.vid)
pprint.pprint(rew)
pprint.pprint(info)