def experiment_setup(args):
if args.vae_dist_help:
load_vaes(args)
#since some extensions of the envs use the distestimator this load is used with the interval wrapper#todo use other?
load_field_parameters(args)
if args.dist_estimator_type is not None:
temp_env = make_temp_env(args)
load_dist_estimator(args, temp_env)
del temp_env
env = make_env(args)
env_test = make_env(args)
if args.goal_based:
args.obs_dims = list(goal_based_process(env.reset()).shape)
args.acts_dims = [env.action_space.shape[0]]
args.compute_reward = env.compute_reward
args.compute_distance = env.compute_distance
if args.imaginary_obstacle_transitions:
#relative small buffer size so it always have most recent collisions
args.imaginary_buffer = ReplayBuffer_Imaginary(
args, buffer_size=args.im_buffer_size)
args.buffer = buffer = ReplayBuffer_Episodic(args)
args.learner = learner = create_learner(args)
args.agent = agent = create_agent(args)
args.logger.info('*** network initialization complete ***')
args.tester = tester = Tester(args)
args.logger.info('*** tester initialization complete ***')
args.timesteps = env.env.env.spec.max_episode_steps
return env, env_test, agent, buffer, learner, tester
def __init__(self, args):
self.args = args
self.env = make_env(args)
self.env_test = make_env(args)
self.env_List = []
for i in range(args.episodes):
self.env_List.append(make_env(args))
self.agent = create_agent(args)
self.achieved_trajectory_pool = TrajectoryPool(args, args.hgg_pool_size)
self.stop_hgg_threshold = self.args.stop_hgg_threshold
self.stop = False
self.learn_calls = 0
def experiment_setup(args):
env = make_env(args)
args.acts_dims = env.acts_dims
args.obs_dims = env.obs_dims
args.buffer = buffer = create_buffer(args)
args.agent = agent = create_agent(args)
args.agent_graph = agent.graph
args.learner = learner = create_learner(args)
args.logger.info('*** network initialization complete ***')
args.tester = tester = Tester(args)
args.logger.info('*** tester initialization complete ***')
return env, agent, buffer, learner, tester
def experiment_setup(args):
env = make_env(args)
env_test = make_env(args)
if args.goal_based:
args.obs_dims = list(goal_based_process(env.reset()).shape)
args.acts_dims = [env.action_space.shape[0]]
args.compute_reward = env.compute_reward
args.compute_distance = env.compute_distance
args.buffer = buffer = ReplayBuffer_Episodic(args)
args.learner = learner = create_learner(args)
args.agent = agent = create_agent(args)
args.logger.info('*** network initialization complete ***')
args.tester = tester = Tester(args)
args.logger.info('*** tester initialization complete ***')
return env, env_test, agent, buffer, learner, tester