base.py

class Environment(object):
  """Base class for environment."""
  def __init__(self):
    pass

  @property
  def num_of_actions(self):
    raise NotImplementedError

  def GetState(self):
    """Returns current state as numpy (possibly) multidimensional array.

    If the current state is terminal, returns None.
    """
    raise NotImplementedError

  def ProcessAction(self, action):
    """Performs one step given selected action. Returns step reward."""
    raise NotImplementedError


class Agent(object):
  """Base class for different agents."""
  def __init__(self):
    pass

  def ChooseAction(self, state):
    pass


def PlayEpisode(env, agent):
  state = env.GetState()
  total_reward = 0.
  while state is not None:
    action = agent.ChooseAction(state)
    total_reward += env.ProcessAction(action)
    state = env.GetState()

  return total_reward