def initPolicy(self, game, agentIndex=0):
"""Interactively allow user to choose an MDP policy."""
from keyboardAgents import KeyboardAgent
from graphicsUtils import keys_pressed
from searchAgents import searchGen
import itertools
agent = game.agents[agentIndex]
def pollChoice(state):
blink_show = True
blink_delay = 3
blink_tick = 0
self.update(state.data)
action = None
while not action:
action = agent.getAction(state)
time.sleep(0.1)
if blink_tick == blink_delay - 1:
blink_show = not blink_show
self.update(state.data, blink_show)
blink_tick = (blink_tick + 1) % blink_delay
return action
wallsDict = game.state.getWalls().asValDict()
availablePos = set(wallsDict[False])
foodDict = game.state.getFood().asPosDict()
delKeys = []
# Eliminate entries that refer to locations without food
for k, v in foodDict.items():
if v in [0, False]:
delKeys.append(k)
for k in delKeys:
del foodDict[k]
minStates = []
for r in range(len(foodDict.items()) + 1):
for foodState in itertools.combinations(foodDict.items(), r):
# import pdb; pdb.set_trace()
print(foodState)
occupiedByFood = set()
for pos, _ in foodState:
occupiedByFood.add(pos)
for pos in availablePos - occupiedByFood:
minStates.append((pos, frozenset(foodState)))
policy = dict()
for s in minStates:
state = GameState()
state.data = game.state.data.deepCopy(
) # Passing data into __init__ makes
# shallow copy instead of deep.
state.data.agentStates[0].configuration.pos = s[0]
state.data.food = Grid(state.data.food.width,
state.data.food.height, 0)
for (x, y), foodVal in s[1]:
state.data.food[x][y] = foodVal
print(s)
policy[s] = pollChoice(state)
self.blinkOn = False
return policy
