Function reward_function

rl3/es_flappy.py:178–210

(params)

def reward_function(params):
  model = ANN(D, M, K)
  model.set_params(params)
  
  # play one episode and return the total reward
  episode_reward = 0
  episode_length = 0 # not sure if it will be used
  done = False
  obs = env.reset()
  obs_dim = len(obs)
  if HISTORY_LENGTH > 1:
    state = np.zeros(HISTORY_LENGTH*obs_dim) # current state
    state[-obs_dim:] = obs
  else:
    state = obs
  while not done:
    # get the action
    action = model.sample_action(state)

    # perform the action
    obs, reward, done = env.step(action)

    # update total reward
    episode_reward += reward
    episode_length += 1

    # update state
    if HISTORY_LENGTH > 1:
      state = np.roll(state, -obs_dim)
      state[-obs_dim:] = obs
    else:
      state = obs
  return episode_reward


if __name__ == '__main__':