def main():
parser = argparse.ArgumentParser()
parser.add_argument(
"--file",
type=str,
default="",
)
args = parser.parse_args()
_, get_action = load_policy(args.file)
env = gym.make('HumanoidRL-v0')
run_policy(env, get_action)
def run(train_fn, env_config, net_config=None):
mode = '--train' if not sys.argv[1:] else sys.argv[1]
if mode == '--train':
train_fn()
elif mode == '--test':
if not sys.argv[2:]:
model_path = get_latest_model_path() # TODO
else:
model_path = sys.argv[2]
env = gym.make(env_config['env_name'])
_, get_action = load_policy_and_env(fpath=model_path, deterministic=True,
env=env, net_config=net_config)
run_policy(env, get_action)
def run(train_fn, env_config, net_config=None):
mode = '--train' if not sys.argv[1:] else sys.argv[1]
if mode == '--train':
train_fn()
elif mode == '--play':
from deepdrive_zero import player
env_config['physics_steps_per_observation'] = 1
player.start(env_config=env_config)
elif mode == '--test':
if not sys.argv[2:]:
model_path = get_latest_model_path() # TODO
else:
model_path = sys.argv[2]
env = gym.make(env_config['env_name'])
env.configure_env(env_config)
_, get_action = load_policy_and_env(fpath=model_path, deterministic=True,
env=env, net_config=net_config)
run_policy(env, get_action)
if TEST_STATIC_OBSTACLE:
_, get_action = load_policy(
'/home/c2/src/spinningup/data/dd2d-ppo-intersection/dd2d-ppo-intersection_s0',
use_model_only=False)
env = gym.make('deepdrive-2d-static-obstacle-no-g-pen-v0')
else:
p = '/home/c2/src/tmp/spinningup/data/deepdrive-2d-intersection-no-constrained-controls-example/deepdrive-2d-intersection-no-constrained-controls-example_s0_2020_03-10_13-14.50/best_HorizonReturn/2020_03-11_11-36.27'
if 'no-end-g' in p or 'no-contraint-g' in p or 'no-g' in p or 'no-constrain' in p:
os.environ['END_ON_HARMFUL_GS'] = '0'
os.environ['GFORCE_PENALTY_COEFF'] = '0'
os.environ['JERK_PENALTY_COEFF'] = '0'
if 'no-constrain' in p:
os.environ['CONSTRAIN_CONTROLS'] = '0'
if 'delta-controls' in p or 'deepdrive-2d-intersection-no-g-or-jerk2' in p:
os.environ['EXPECT_NORMALIZED_ACTION_DELTAS'] = '1'
else:
os.environ['EXPECT_NORMALIZED_ACTION_DELTAS'] = '0'
if 'one-waypoint' in p:
env_name = 'deepdrive-2d-one-waypoint-v0'
else:
env_name = 'deepdrive-2d-intersection-w-gs-allow-decel-v0'
_, get_action = load_policy_and_env(p, deterministic=True)
# env = gym.make('deepdrive-2d-intersection-v0')
env = gym.make(env_name)
env.configure_env()
# env.unwrapped.physics_steps_per_observation = 1
run_policy(env, get_action)
def test_ppo(exp_dir, itr='last'):
_, get_action, lstm = load_policy_and_env(exp_dir, itr=itr)
env = hide_and_seek.make_env()
run_policy(env, get_action, lstm=lstm)
import gym import spinup import tensorflow as tf import spinup.utils.logx from spinup.utils.test_policy import load_policy, run_policy ''' # METHOD 1 (See Docs) # BipedalWalker-v2 modelpath = '/home/watate/[email protected]/Python/spinningup/data/ex5_ddpg_100ep_bipedalwalker-v2/ex5_ddpg_100ep_bipedalwalker-v2_s0' len = 0 episodes = 100 norender = False #itr = -1 itr = -100 #Only for soft-actor critic deterministic = False # This part is unecessary because load_policy already restores tf_graph #model = spinup.utils.logx.restore_tf_graph(sess, modelpath) env, get_action = load_policy(modelpath, itr if itr >=0 else 'last', deterministic) run_policy(env, get_action, len, episodes, not(norender)) '''
from spinup.utils.test_policy import load_policy, run_policy
import gym
import gym_geofriend2
_, get_action = load_policy('./spinupPpo')
env = gym.make("geofriend2-v0") #Pyramid(),
run_policy(env, get_action, max_ep_len=200)
from spinup.utils.test_policy import load_policy, run_policy
import gym
import argparse
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--env', type=str, required=True)
parser.add_argument('--save_dir', type=str, required=True)
parser.add_argument("-d", '--deterministic', action="store_true")
parser.add_argument("-e", "--episodes", type=int, default=100)
parser.add_argument("-dr", "--dont_render", action="store_false")
args = parser.parse_args()
if args.dont_render == False:
render_it = False
else:
render_it = True
print(args.save_dir)
_, get_action = load_policy(args.save_dir,
deterministic=args.deterministic)
env = gym.make(args.env)
run_policy(env, get_action, num_episodes=args.episodes, render=render_it)