# your_best.py
# Food ROI targeting candidate for CS470 Assignment 3.

from captureAgents import CaptureAgent
import random
import util
from game import Directions
from util import nearestPoint


def createTeam(firstIndex, secondIndex, isRed,
               first='FoodROIOffensiveAgent',
               second='FoodROISupportAgent'):
  return [eval(first)(firstIndex), eval(second)(secondIndex)]


class ReflexCaptureAgent(CaptureAgent):
  def registerInitialState(self, gameState):
    self.start = gameState.getAgentPosition(self.index)
    self.width = gameState.data.layout.width
    self.height = gameState.data.layout.height
    CaptureAgent.registerInitialState(self, gameState)
    self.homeEntries = self.getHomeEntries(gameState)

  def chooseAction(self, gameState):
    actions = gameState.getLegalActions(self.index)
    values = [self.evaluate(gameState, action) for action in actions]
    bestValue = max(values)
    bestActions = [a for a, v in zip(actions, values) if v == bestValue]
    if Directions.STOP in bestActions and len(bestActions) > 1:
      bestActions.remove(Directions.STOP)
    return random.choice(bestActions)

  def getSuccessor(self, gameState, action):
    successor = gameState.generateSuccessor(self.index, action)
    pos = successor.getAgentState(self.index).getPosition()
    if pos != nearestPoint(pos):
      return successor.generateSuccessor(self.index, action)
    return successor

  def evaluate(self, gameState, action):
    return self.getFeatures(gameState, action) * self.getWeights(gameState, action)

  def getFeatures(self, gameState, action):
    features = util.Counter()
    successor = self.getSuccessor(gameState, action)
    features['successorScore'] = self.getScore(successor)
    return features

  def getWeights(self, gameState, action):
    return {'successorScore': 1.0}

  def getHomeEntries(self, gameState):
    walls = gameState.getWalls()
    x = self.width // 2 - 1 if self.red else self.width // 2
    entries = []
    for y in range(1, self.height - 1):
      if not walls[x][y]:
        entries.append((x, y))
    return entries or [self.start]

  def minDistance(self, pos, targets):
    if not pos or not targets:
      return 0
    return min(self.getMazeDistance(pos, target) for target in targets)

  def activeEnemyGhosts(self, gameState):
    ghosts = []
    for opponent in self.getOpponents(gameState):
      enemy = gameState.getAgentState(opponent)
      enemyPos = enemy.getPosition()
      if enemyPos is None:
        continue
      if not enemy.isPacman and enemy.scaredTimer <= 1:
        ghosts.append((opponent, enemyPos))
    return ghosts

  def activeEnemyGhostDistances(self, gameState, myPos):
    return [self.getMazeDistance(myPos, pos)
            for opponent, pos in self.activeEnemyGhosts(gameState)]


class FoodROIOffensiveAgent(ReflexCaptureAgent):
  def registerInitialState(self, gameState):
    ReflexCaptureAgent.registerInitialState(self, gameState)
    self.currentFoodTarget = None
    self.currentReturnMode = False

  def chooseAction(self, gameState):
    myState = gameState.getAgentState(self.index)
    myPos = myState.getPosition()
    foodList = self.getFood(gameState).asList()
    homeDistance = self.minDistance(myPos, self.homeEntries)
    ghostDistances = self.activeEnemyGhostDistances(gameState, myPos)
    closestGhost = min(ghostDistances) if ghostDistances else None

    self.currentFoodTarget = self.bestFoodTarget(gameState, myPos, foodList)
    self.currentReturnMode = (
      myState.numCarrying >= 3 or
      len(foodList) <= 2 or
      (myState.numCarrying > 0 and closestGhost is not None and closestGhost <= 5) or
      gameState.data.timeleft < homeDistance + 20
    )
    return ReflexCaptureAgent.chooseAction(self, gameState)

  def bestFoodTarget(self, gameState, myPos, foodList):
    if not myPos or not foodList:
      return None

    clusterBonuses = self.foodClusterBonuses(foodList)
    ghosts = self.activeEnemyGhosts(gameState)
    bestFood = None
    bestCost = None

    for food in foodList:
      myDistance = self.getMazeDistance(myPos, food)
      homeDistance = self.minDistance(food, self.homeEntries)
      cost = (
        myDistance +
        0.6 * homeDistance +
        self.ghostRiskPenalty(food, myDistance, ghosts) -
        clusterBonuses.get(food, 0.0) +
        self.teammateOverlapPenalty(gameState, food, myDistance)
      )

      if bestCost is None or cost < bestCost:
        bestCost = cost
        bestFood = food
      elif cost == bestCost and bestFood is not None:
        if myDistance < self.getMazeDistance(myPos, bestFood):
          bestFood = food

    return bestFood

  def foodClusterBonuses(self, foodList):
    bonuses = {}
    for food in foodList:
      bonus = 0.0
      for other in foodList:
        if other == food:
          continue
        distance = abs(food[0] - other[0]) + abs(food[1] - other[1])
        if distance <= 2:
          bonus += 2.0
        elif distance <= 4:
          bonus += 1.0
        elif distance <= 6:
          bonus += 0.4
      bonuses[food] = min(7.0, bonus)
    return bonuses

  def ghostRiskPenalty(self, food, myDistance, ghosts):
    if not ghosts:
      return 0.0

    penalty = 0.0
    for opponent, ghostPos in ghosts:
      ghostDistance = self.getMazeDistance(ghostPos, food)
      margin = ghostDistance - myDistance
      if margin <= 0:
        penalty += 16.0 + min(10.0, abs(margin) * 2.0)
      elif margin <= 2:
        penalty += 9.0 - 2.0 * margin
      elif margin <= 4:
        penalty += 2.5
    return min(35.0, penalty)

  def teammateOverlapPenalty(self, gameState, food, myDistance):
    penalty = 0.0
    for teammate in self.getTeam(gameState):
      if teammate == self.index:
        continue
      teammatePos = gameState.getAgentPosition(teammate)
      if teammatePos is None:
        continue
      teammateDistance = self.getMazeDistance(teammatePos, food)
      if teammateDistance + 1 < myDistance:
        penalty += 8.0
      elif teammateDistance <= myDistance + 1:
        penalty += 3.0

    team = sorted(self.getTeam(gameState))
    if len(team) >= 2:
      midpoint = self.height / 2.0
      lowerLaneAgent = team[0]
      if self.index == lowerLaneAgent and food[1] > midpoint:
        penalty += 1.5
      elif self.index != lowerLaneAgent and food[1] < midpoint:
        penalty += 1.5
    return penalty

  def getFeatures(self, gameState, action):
    features = util.Counter()
    successor = self.getSuccessor(gameState, action)
    myState = successor.getAgentState(self.index)
    myPos = myState.getPosition()

    foodList = self.getFood(successor).asList()
    capsules = self.getCapsules(successor)
    carrying = myState.numCarrying
    ghostDistances = self.activeEnemyGhostDistances(successor, myPos)
    closestGhost = min(ghostDistances) if ghostDistances else None

    features['successorScore'] = self.getScore(successor)
    features['foodRemaining'] = len(foodList)
    if self.currentFoodTarget is not None:
      features['distanceToFood'] = self.getMazeDistance(myPos, self.currentFoodTarget)
    else:
      features['distanceToFood'] = self.minDistance(myPos, foodList)
    features['distanceHome'] = self.minDistance(myPos, self.homeEntries)
    features['distanceToCapsule'] = self.minDistance(myPos, capsules)

    if action == Directions.STOP:
      features['stop'] = 1
    reverse = Directions.REVERSE[gameState.getAgentState(self.index).configuration.direction]
    if action == reverse:
      features['reverse'] = 1

    if closestGhost is not None:
      features['ghostDistance'] = closestGhost
      if closestGhost <= 2:
        features['immediateDanger'] = 1
      elif closestGhost <= 5:
        features['nearGhost'] = 1

    if carrying >= 3 or len(foodList) <= 2:
      features['shouldReturn'] = 1
    if carrying > 0 and closestGhost is not None and closestGhost <= 5:
      features['shouldReturn'] = 1
    if gameState.data.timeleft < features['distanceHome'] + 20:
      features['shouldReturn'] = 1

    return features

  def getWeights(self, gameState, action):
    weights = {
      'successorScore': 200,
      'foodRemaining': -100,
      'distanceToFood': -3,
      'distanceToCapsule': -2,
      'ghostDistance': 2,
      'immediateDanger': -1000,
      'nearGhost': -180,
      'stop': -100,
      'reverse': -3,
      'shouldReturn': 0,
      'distanceHome': 0,
    }
    if self.currentReturnMode:
      weights['distanceHome'] = -15
      weights['distanceToFood'] = -1
      weights['distanceToCapsule'] = -1
    return weights


class FoodROISupportAgent(FoodROIOffensiveAgent):
  def registerInitialState(self, gameState):
    FoodROIOffensiveAgent.registerInitialState(self, gameState)
    self.lastEatenFood = None
    self.patrolTarget = self.choosePatrolTarget(gameState)
    self.supportEnabled = self.height > 7

  def chooseAction(self, gameState):
    self.updateLastEatenFood(gameState)
    myState = gameState.getAgentState(self.index)
    invaders = self.visibleInvaders(gameState)

    if invaders and not myState.isPacman:
      myPos = gameState.getAgentPosition(self.index)
      target = min(
        [invader.getPosition() for invader in invaders],
        key=lambda p: self.getMazeDistance(myPos, p)
      )
      return self.chooseDefensiveMove(gameState, target, chase=True)

    if self.lastEatenFood is not None and not myState.isPacman and self.getScore(gameState) >= 1:
      return self.chooseDefensiveMove(gameState, self.lastEatenFood, chase=False)

    if not self.supportEnabled:
      return self.chooseDefensiveMove(gameState, self.patrolTarget, chase=False)

    if myState.numCarrying == 0 and self.getScore(gameState) < 1:
      return self.chooseDefensiveMove(gameState, self.patrolTarget, chase=False)

    return FoodROIOffensiveAgent.chooseAction(self, gameState)

  def chooseDefensiveMove(self, gameState, target, chase):
    actions = gameState.getLegalActions(self.index)
    values = [self.defensiveMoveScore(gameState, action, target, chase) for action in actions]
    bestValue = max(values)
    bestActions = [action for action, value in zip(actions, values) if value == bestValue]
    if Directions.STOP in bestActions and len(bestActions) > 1:
      bestActions.remove(Directions.STOP)
    return random.choice(bestActions)

  def defensiveMoveScore(self, gameState, action, target, chase):
    successor = self.getSuccessor(gameState, action)
    myState = successor.getAgentState(self.index)
    myPos = myState.getPosition()
    score = 0.0

    if myState.isPacman:
      score -= 1000.0
    else:
      score += 120.0

    if target is not None:
      weight = 25.0 if chase else 4.0
      score -= weight * self.getMazeDistance(myPos, target)

    if action == Directions.STOP:
      score -= 100.0
    reverse = Directions.REVERSE[gameState.getAgentState(self.index).configuration.direction]
    if action == reverse:
      score -= 3.0
    return score

  def visibleInvaders(self, gameState):
    enemies = [gameState.getAgentState(i) for i in self.getOpponents(gameState)]
    return [enemy for enemy in enemies if enemy.isPacman and enemy.getPosition() is not None]

  def updateLastEatenFood(self, gameState):
    previous = self.getPreviousObservation()
    if previous is None:
      return
    previousFood = set(self.getFoodYouAreDefending(previous).asList())
    currentFood = set(self.getFoodYouAreDefending(gameState).asList())
    eaten = list(previousFood - currentFood)
    if eaten:
      myPos = gameState.getAgentPosition(self.index)
      self.lastEatenFood = min(eaten, key=lambda p: self.getMazeDistance(myPos, p))
    elif self.lastEatenFood is not None:
      myPos = gameState.getAgentPosition(self.index)
      if myPos is not None and self.getMazeDistance(myPos, self.lastEatenFood) <= 1:
        self.lastEatenFood = None

  def choosePatrolTarget(self, gameState):
    defendingFood = self.getFoodYouAreDefending(gameState).asList()
    if not defendingFood:
      return random.choice(self.homeEntries)
    foodCenter = (
      sum(x for x, y in defendingFood) / float(len(defendingFood)),
      sum(y for x, y in defendingFood) / float(len(defendingFood))
    )
    return min(
      self.homeEntries,
      key=lambda p: abs(p[0] - foodCenter[0]) + abs(p[1] - foodCenter[1])
    )


class PatrolDefensiveAgent(ReflexCaptureAgent):
  def registerInitialState(self, gameState):
    ReflexCaptureAgent.registerInitialState(self, gameState)
    self.patrolTarget = self.choosePatrolTarget(gameState)
    self.lastEatenFood = None

  def chooseAction(self, gameState):
    self.updateLastEatenFood(gameState)
    invaders = self.visibleInvaders(gameState)
    if invaders:
      myPos = gameState.getAgentPosition(self.index)
      self.patrolTarget = min(
        [a.getPosition() for a in invaders],
        key=lambda p: self.getMazeDistance(myPos, p)
      )
    elif self.lastEatenFood is not None:
      self.patrolTarget = self.lastEatenFood
    elif self.patrolTarget is None:
      self.patrolTarget = self.choosePatrolTarget(gameState)
    return ReflexCaptureAgent.chooseAction(self, gameState)

  def getFeatures(self, gameState, action):
    features = util.Counter()
    successor = self.getSuccessor(gameState, action)
    myState = successor.getAgentState(self.index)
    myPos = myState.getPosition()

    invaders = self.visibleInvaders(successor)
    features['onDefense'] = 1
    if myState.isPacman:
      features['onDefense'] = 0
    features['numInvaders'] = len(invaders)
    if invaders:
      features['invaderDistance'] = min(
        self.getMazeDistance(myPos, invader.getPosition()) for invader in invaders
      )
    else:
      features['distanceToPatrol'] = self.getMazeDistance(myPos, self.patrolTarget)

    if action == Directions.STOP:
      features['stop'] = 1
    reverse = Directions.REVERSE[gameState.getAgentState(self.index).configuration.direction]
    if action == reverse:
      features['reverse'] = 1
    return features

  def getWeights(self, gameState, action):
    return {
      'numInvaders': -1000,
      'onDefense': 120,
      'invaderDistance': -20,
      'distanceToPatrol': -4,
      'stop': -100,
      'reverse': -2,
    }

  def visibleInvaders(self, gameState):
    enemies = [gameState.getAgentState(i) for i in self.getOpponents(gameState)]
    return [enemy for enemy in enemies if enemy.isPacman and enemy.getPosition() is not None]

  def updateLastEatenFood(self, gameState):
    previous = self.getPreviousObservation()
    if previous is None:
      return
    previousFood = set(self.getFoodYouAreDefending(previous).asList())
    currentFood = set(self.getFoodYouAreDefending(gameState).asList())
    eaten = list(previousFood - currentFood)
    if eaten:
      myPos = gameState.getAgentPosition(self.index)
      self.lastEatenFood = min(eaten, key=lambda p: self.getMazeDistance(myPos, p))

  def choosePatrolTarget(self, gameState):
    defendingFood = self.getFoodYouAreDefending(gameState).asList()
    if not defendingFood:
      return random.choice(self.homeEntries)
    foodCenter = (
      sum(x for x, y in defendingFood) / float(len(defendingFood)),
      sum(y for x, y in defendingFood) / float(len(defendingFood))
    )
    return min(
      self.homeEntries,
      key=lambda p: abs(p[0] - foodCenter[0]) + abs(p[1] - foodCenter[1])
    )