def main(args):
render = args.render
if not render:
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
from utils.utils import TabularPolicy, TabularValueFun
from part1.tabular_value_iteration import ValueIteration
from envs import Grid1DEnv, GridWorldEnv
envs = [GridWorldEnv(seed=0), GridWorldEnv(seed=1)]
for env in envs:
env_name = env.__name__
exp_dir = os.getcwd() + '/data/part1/%s/policy_type%s_temperature%s/' % (env_name, args.policy_type, args.temperature)
logger.configure(dir=exp_dir, format_strs=['stdout', 'log', 'csv'])
args_dict = vars(args)
args_dict['env'] = env_name
json.dump(vars(args), open(exp_dir + '/params.json', 'w'), indent=2, sort_keys=True)
policy = TabularPolicy(env)
value_fun = TabularValueFun(env)
algo = ValueIteration(env,
value_fun,
policy,
policy_type=args.policy_type,
render=render,
temperature=args.temperature)
algo.train()
def main(args):
render = args.render
if not render:
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
from envs import DoubleIntegratorEnv, MountainCarEnv, CartPoleEnv, SwingUpEnv
from utils.utils import TabularPolicy, TabularValueFun
from part1.tabular_value_iteration import ValueIteration
from part2.look_ahead_policy import LookAheadPolicy
from part2.discretize import Discretize
envs = [
DoubleIntegratorEnv(),
MountainCarEnv(),
CartPoleEnv(),
SwingUpEnv()
]
for env in envs:
env_name = env.__class__.__name__
if env_name == 'MountainCarEnv':
state_discretization = 51
else:
state_discretization = 21
exp_dir = os.getcwd(
) + '/data/part2_d/%s/policy_type%s_mode%s_horizon%s/' % (
env_name, args.policy_type, args.mode, args.horizon)
logger.configure(dir=exp_dir, format_strs=['stdout', 'log', 'csv'])
args_dict = vars(args)
args_dict['env'] = env_name
json.dump(vars(args),
open(exp_dir + '/params.json', 'w'),
indent=2,
sort_keys=True)
env = Discretize(env,
state_discretization=state_discretization,
mode=args.mode)
value_fun = TabularValueFun(env)
if args.policy_type == 'tabular':
policy = TabularPolicy(env)
elif args.policy_type == 'look_ahead':
policy = LookAheadPolicy(env, value_fun, args.horizon)
else:
raise NotImplementedError
algo = ValueIteration(env,
value_fun,
policy,
render=render,
max_itr=args.max_iter,
num_rollouts=1,
render_itr=5,
log_itr=5)
algo.train()
def main(args):
render = args.render
if not render:
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
from utils.utils import TabularPolicy
from utils.value_function import TabularValueFun
from algos.tabular_value_iteration import ValueIteration
from envs import ASRSEnv, TabularEnv, ProbDistEnv, DynamicProbEnv, StaticOrderProcess, SeasonalOrderProcess
num_products = np.array(eval(args.storage_shape)).prod()
assert (eval(args.dist_param) is None) or (num_products == len(
eval(args.dist_param)
)), 'storage_shape should be consistent with dist_param length'
op = StaticOrderProcess(num_products=num_products,
dist_param=eval(args.dist_param))
base_env = ASRSEnv(eval(args.storage_shape),
order_process=op,
origin_coord=eval(args.exit_coord))
env = TabularEnv(base_env)
env_name = env.__name__
exp_dir = os.getcwd(
) + '/data/version1/%s/policy_type%s_temperature%s_envsize_%s/' % (
env_name, args.policy_type, args.temperature,
np.array(eval(args.storage_shape)).prod())
logger.configure(dir=exp_dir,
format_strs=['stdout', 'log', 'csv'],
level=eval(args.logger_level))
args_dict = vars(args)
args_dict['env'] = env_name
json.dump(vars(args),
open(exp_dir + '/params.json', 'w'),
indent=2,
sort_keys=True)
policy = TabularPolicy(env)
value_fun = TabularValueFun(env)
algo = ValueIteration(env,
value_fun,
policy,
policy_type=args.policy_type,
render=render,
temperature=args.temperature,
num_rollouts=args.num_rollouts)
algo.train()
value_fun.save(f'{exp_dir}/value_fun.npy')
def main(args):
render = args.render
if not render:
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
from envs import CartPoleEnv, SwingUpEnv
from utils.utils import TabularPolicy, TabularValueFun
from part1.tabular_value_iteration import ValueIteration
from part4.discretize import Discretize
envs = [CartPoleEnv(), SwingUpEnv()]
for env in envs:
env_name = env.__class__.__name__
exp_dir = os.getcwd(
) + '/data/part4/%s/mode%s_state_discretization%s/' % (
env_name, args.mode, str(args.state_discretization))
logger.configure(dir=exp_dir, format_strs=['stdout', 'log', 'csv'])
args_dict = vars(args)
args_dict['env'] = env_name
json.dump(vars(args),
open(exp_dir + '/params.json', 'w'),
indent=2,
sort_keys=True)
env = Discretize(env,
state_discretization=args.state_discretization,
mode=args.mode)
value_fun = TabularValueFun(env)
policy = TabularPolicy(env)
algo = ValueIteration(env,
value_fun,
policy,
render=render,
max_itr=args.max_iter,
num_rollouts=1,
render_itr=5,
log_itr=5)
algo.train()