def main(folderName): # folderName = './log-files/SecondModel/Jan-30_20_38_36' # chkp.print_tensors_in_checkpoint_file("{}/value_function.ckpt".format(folderName), tensor_name='', all_tensors=True) env = myEnv() obs_dim = len(env.observation_space()) act_dim = len(env.action_space()) obs_dim += 1 # add 1 to obs dimension for time step feature (see run_episode()) policy = Policy(obs_dim, act_dim, 0.003) policy.restore(folderName) trajectories = run_policy(env, policy, episodes=20)
def init_env(env_name): ''' Inicialize my environment and return dimension of observation and action spaces. Args: Env_name: str environment name (e.g. "SecondModel") Returns: 3-tuple Environment (object) Number of observation dimension (int) Number of action dimensions (int) ''' env = myEnv() obs_dim = len(env.observation_space()) act_dim = len(env.action_space()) return env, obs_dim, act_dim
def init_gym(env_name): """ Initialize gym environment, return dimension of observation and action spaces. Args: env_name: str environment name (e.g. "Humanoid-v1") Returns: 3-tuple gym environment (object) number of observation dimensions (int) number of action dimensions (int) """ #env = gym.make(env_name) env = myEnv() obs_dim = len(env.observation_space()) act_dim = len(env.action_space()) return env, obs_dim, act_dim
def main(folderName, bestMode): # folderName = './log-files/SecondModel/Jan-30_20_38_36' # chkp.print_tensors_in_checkpoint_file("{}/value_function.ckpt".format(folderName), tensor_name='', all_tensors=True) env = myEnv() if not bestMode: obs_dim = len(env.observation_space()) act_dim = len(env.action_space()) obs_dim += 1 # add 1 to obs dimension for time step feature (see run_episode()) scaler = Scaler(obs_dim) policy = Policy(obs_dim, act_dim, 0.003) policy.restore(folderName) trajectories = run_policy(env, policy, scaler, episodes=20) else: sequences = recoverSequences(folderName) for sequence in sequences: runSeq(sequence['actions'], env)