argument.py

def add_arguments(parser):

    parser.add_argument('--frame_width', type=int, default = 84, help='Resized frame width')
    parser.add_argument('--frame_height', type=int, default = 84, help='Resized frame height')
    parser.add_argument('--num_steps', type=int, default = 10000000, help='Number of episodes the agent plays')
    parser.add_argument('--state_length', type=int, default = 4, help='Number of most recent frames to produce the input to the network')
    parser.add_argument('--gamma', type=float, default = 0.99, help='Discount factor')
    parser.add_argument('--exploration_steps', type=int, default = 1000000, help='Number of steps over which the initial value of epsilon is linearly annealed to its final value')
    parser.add_argument('--initial_epsilon', type=float, default = 1.0, help='Initial value of epsilon in epsilon-greedy')
    parser.add_argument('--final_epsilon', type=float, default = 0.05, help='Final value of epsilon in epsilon-greedy')
    parser.add_argument('--initial_replay_size', type=int, default = 10000, help='Number of steps to populate the replay memory before training starts')
    parser.add_argument('--num_replay_memory', type=int, default = 10000, help='Number of replay memory the agent uses for training')
    parser.add_argument('--batch_size', type=int, default = 32, help='Mini batch size')
    parser.add_argument('--target_update_interval', type=int, default = 1000, help='The frequency with which the target network is updated')
    parser.add_argument('--train_interval', type=int, default = 4, help='The agent selects 4 actions between successive updates')
    parser.add_argument('--learning_rate', type=float, default = 0.0001, help='Learning rate used by RMSProp')
    parser.add_argument('--min_grad', type=float, default = 1e-8, help='Constant added to the squared gradient in the denominator of the RMSProp update')
    parser.add_argument('--save_interval', type=int, default = 300000, help='The frequency with which the network is saved')
    parser.add_argument('--no_op_steps', type=int, default = 2, help='Maximum number of "do nothing" actions to be performed by the agent at the start of an episode')
    parser.add_argument('--save_network_path', type=str, default = "saved_dqn_networks/", help='')
    parser.add_argument('--save_summary_path', type=str, default = "dqn_summary/", help='')
    parser.add_argument('--test_dqn_model_path', type=str, default = "pretrained_model/dqn_300.h5", help='')
    parser.add_argument('--exp_name', type=str, default = "breakout_dqn", help='')
    parser.add_argument('--gpu_frac', type=float, default = 1.0, help='Set GPU use limit for tensorflow')
    parser.add_argument('--ddqn', type=bool, default = False, help='Set True to apply Double Q-learning')
    parser.add_argument('--dueling', type=bool, default = False, help='Set True to apply Duelinng Network')
    parser.add_argument('--test_path', type=str, default = 'saved_dqn_networks/breakout_mse7200000.h5', help='Model used during testing')
    parser.add_argument('--actor_critic',type=bool, default= False, help='Set True to use actor critc model')
    parser.add_argument('--optimizer',type=str, default='rmsprop', help='Optimizer (Adam or Rmsp)')

    return parser