Function play_one

rl2/mountaincar/td_lambda.py:78–98

(model, env, eps, gamma, lambda_)

# returns a list of states_and_rewards, and the total reward
def play_one(model, env, eps, gamma, lambda_):
  observation = env.reset()
  done = False
  totalreward = 0
  iters = 0
  # while not done and iters < 200:
  while not done and iters < 10000:
    action = model.sample_action(observation, eps)
    prev_observation = observation
    observation, reward, done, info = env.step(action)

    # update the model
    Qnext = model.predict(observation)
    assert(Qnext.shape == (1, env.action_space.n))
    G = reward + gamma*np.max(Qnext[0])
    model.update(prev_observation, action, G, gamma, lambda_)

    totalreward += reward
    iters += 1

  return totalreward


if __name__ == '__main__':