How to use the accountant.Accountant function in accountant

def main():
    # parse arguments
    parser = argparse.ArgumentParser()
    parser.add_argument('--agent', type=str, default='GAMBLER', help='Which agent to use')
    parser.add_argument('--learning_rate', type=float, default=0.5, help='How quickly the algorithm tries to learn')
    parser.add_argument('--discount', type=float, default=0.98, help='Discount for estimated future action')
    parser.add_argument('--iterations', type=int, default=2000, help='Iteration count')
    ARGS, unparsed = parser.parse_known_args()

    # select agent
    if ARGS.agent == 'GAMBLER':
        agent = Gambler(learning_rate=ARGS.learning_rate, discount=ARGS.discount, iterations=ARGS.iterations)
    elif ARGS.agent == 'ACCOUNTANT':
        agent = Accountant()
    elif ARGS.agent == 'DEEPGAMBLER':
        agent = DeepGambler(learning_rate=ARGS.learning_rate, discount=ARGS.discount, iterations=ARGS.iterations)
    else:
        agent = Drunkard()

    # setup simulation
    dungeon = DungeonSimulator()
    dungeon.reset()
    total_reward = 0 # Score keeping
    last_total = 0

    # main loop
    for step in range(ARGS.iterations):
        old_state = dungeon.state # Store current state
        action = agent.get_next_action(old_state) # Query agent for the next action
        new_state, reward = dungeon.take_action(action) # Take action, get new state and reward