Question

The world contains agents at different locations, with only a single agent at any location. Each agent knows where he’s at, but I also need to quickly check if there’s an agent at a given location. Hence, I also maintain a map from locations to agents. I have a problem deciding where this map belongs to: class World, class Agent (as a class attribute) or elsewhere.

In the following I put the lookup table, agent_locations, in class World. But now agents have to call world.update_agent_location every time they move. This is very annoying; what if I decide later to track other things about the agents, apart from their locations – would I need to add calls back to the world object all across the Agent code?

class World:
  def __init__(self, n_agents):
    # ...
    self.agents = []
    self.agent_locations = {}
    for id in range(n_agents):
      x, y = self.find_location()
      agent = Agent(self,x,y)
      self.agents.append(agent)
      self.agent_locations[x,y] = agent
  def update_agent_location(self, agent, x, y):
    del self.agent_locations[agent.x, agent.y]
    self.agent_locations[x, y] = agent
  def update(self): # next step in the simulation
    for agent in self.agents:
      agent.update() # next step for this agent
  # ...

class Agent:
  def __init__(self, world, x, y):
    self.world = world
    self.x, self.y = x, y
  def move(self, x1, y1):
    self.world.update_agent_location(self, x1, y1)
    self.x, self.y = x1, y1
  def update():
    # find a good location that is not occupied and move there
    for x, y in self.valid_locations():
      if not self.location_is_good(x, y):
        continue
      if self.world.agent_locations[x, y]: # location occupied
        continue
      self.move(x, y)

I can instead put agent_locations in class Agent as a class attribute. But that only works when I have a single World object. If I later decide to instantiate multiple World objects, the lookup tables would need to be world-specific.

I am sure there’s a better solution…

EDIT: I added a few lines to the code to show how agent_locations is used. Note that it’s only used from inside Agent objects, but I don’t know if that would remain the case forever.

Answer 1

It helps with OOP to talk about the objects in terms of is a and has a. A World has a list of Agents and a list of Locations. A Location has an Agent. An Agent has a Location and a World.

class Agent:
    def __init__(self, world):
        self.location = None
        self.world = world

    def move(self, new_location):
        if self.location is not None:
            self.location.agent = None
        new_location.agent = self
        self.location = new_location

    def update(self):
        for new_location in self.world.locations:
            if self.location_is_good(new_location):
                self.move(new_location)

    def location_is_good(self, location):
        if location.agent is not None:
            return False

class Location:
    def __init__(self, x, y):
        self.x = x
        self.y = y
        self.agent = None

Go through the mental exercise of adding a new attribute to a Location, such as terrain, and it’s easy to see the benefits of such encapsulation. Likewise, adding new things to Agent, such as a weapon, just requires a weapon-specific function similar to move(). Note that the World doesn’t need to get involved in the move() at all. The move is handled strictly between the Agent and the Location.