| import argparse |
|
|
|
|
| def get_parser(): |
| env_types = ["office", "water", "cheetah", "toy", "taxi"] |
|
|
| parser = argparse.ArgumentParser(prog="run_experiments", description='Runs RL experiment over a particular environment using reward shaping based on LTL') |
| |
| |
| parser.add_argument('--reward_types', default="p", type=str, |
| help='This parameter indicated reward types. The options are: p, d, h, which means p: progress, d: distance, h: hybrid') |
| parser.add_argument('--use_adrs', default=False, type=bool, |
| help='This parameter indicated using adaptive reward shaping or not') |
| parser.add_argument('--hybrid_eta', default=0.005, type=float, |
| help='This parameter indicated adaptive reward shaping gamma factor for trade offs between distance and progress') |
| parser.add_argument('--adrs_update', default=500, type=int, |
| help='This parameter indicated update step for distance function adaptive reward shaping') |
| parser.add_argument('--adrs_mu', default=0.5, type=float, |
| help='adrs mu for ADRS') |
| parser.add_argument('--human', default=False, type=bool, |
| help='reward desgined by human e.x. extrinc reward') |
| parser.add_argument('--version', default=1, type=int, |
| help='different type of reward function update 0: update reward function with trajectories 1: update reward function with the best..') |
| parser.add_argument('--theta', default=-1, type=int, |
| help='theta for adrs update') |
| |
| |
| parser.add_argument('--env_name', default='office', type=str, |
| help='This parameter indicated which env types to solve. The options are: ' + str(env_types)) |
| parser.add_argument('--default_setting', default=True, type=bool, |
| help='use default setting') |
| parser.add_argument('--episode_step', default=int(1e2), type=int, |
| help='This parameter indicated how many steps are allowed in one episode') |
| parser.add_argument('--noise_level', default=0, type=float, |
| help='noise level for action') |
| parser.add_argument('--gamma', default=0.9, type=float, |
| help='This parameter indicated MDP discount factor') |
| parser.add_argument('--violation_end', default=False, type=bool, |
| help='Environment will end if the designed specification is violated') |
| parser.add_argument('--missing', default=False, type=bool, |
| help='Environment will have no goal state') |
| parser.add_argument('--map_size', default=1, type=int, |
| help='This parameter indicated especially for office world, increase map size') |
| parser.add_argument('--map_id', default=1, type=int, |
| help='This parameter will be used to decide which map we will use ') |
| |
| |
| parser.add_argument('--eval_freq', default=100, type=int, |
| help='This parameter indicated evaluation frequncy') |
| parser.add_argument('--seed', default=42, type=int, |
| help='default seed value') |
| parser.add_argument('--rolling', default=20, type=int, |
| help='rolling window for evaluation') |
| |
| |
| parser.add_argument('--algo_name', default="dqn", type=str, |
| help='This parameter indicated which algorithm for RL we will use') |
| parser.add_argument('--total_timesteps', default=int(1e4), type=int, |
| help='This parameter indicated total training time steps') |
| parser.add_argument('--total_run', default=3, type=int, |
| help='This parameter indicated how many times we will run each environment with different seeds') |
| parser.add_argument('--use_one_hot', default=False, type=bool, |
| help='This parameter indicated whether we will use one hot encoding or not') |
| parser.add_argument('--node_embedding', default=False, type=bool, |
| help='This parameter indicated whether we will use qrm like representation') |
| parser.add_argument('--lr_start', default=0.1, type=float, |
| help='This parameter will learning rate start') |
| parser.add_argument('--lr_end', default=0.0001, type=float, |
| help='This parameter will learnint rate end') |
| parser.add_argument('--lr_fraction', default=0.4, type=float, |
| help='This parameter will learning rate fraction') |
| parser.add_argument('--exp_init', default=0.1, type=float, |
| help='This parameter will exploration rate start') |
| parser.add_argument('--exp_final', default=0.05, type=float, |
| help='This parameter will exploration rate end') |
| parser.add_argument('--exp_fraction', default=0.4, type=float, |
| help='This parameter will exploration fraction') |
| parser.add_argument('--batch_size', default=32, type=int, |
| help='batch size') |
| parser.add_argument('--buffer_size', default=32, type=int, |
| help='buffer size') |
| parser.add_argument('--train_freq', default=1, type=int, |
| help='train_freq') |
| parser.add_argument('--gradient_steps', default=1, type=int, |
| help='gradient steps') |
| parser.add_argument('--update_interval', default=100, type=int, |
| help='update interval') |
| parser.add_argument('--learning_starts', default=1, type=int, |
| help='when we start to train algorithm for off-policy methods') |
| parser.add_argument('--max_grad_norm', default=1., type=float, |
| help='maximum value for gradient norm') |
| |
| |
| parser.add_argument('--hp_tuning', default=False, type=bool, |
| help='hyperparam tuning') |
| args = parser.parse_args() |
|
|
| return args |
|
|
| |
|
|
