-
Notifications
You must be signed in to change notification settings - Fork 0
/
multiAgents.py
402 lines (307 loc) · 16.3 KB
/
multiAgents.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
# multiAgents.py
# --------------
# Licensing Information: You are free to use or extend these projects for
# educational purposes provided that (1) you do not distribute or publish
# solutions, (2) you retain this notice, and (3) you provide clear
# attribution to UC Berkeley, including a link to http://ai.berkeley.edu.
#
# Attribution Information: The Pacman AI projects were developed at UC Berkeley.
# The core projects and autograders were primarily created by John DeNero
# ([email protected]) and Dan Klein ([email protected]).
# Student side autograding was added by Brad Miller, Nick Hay, and
# Pieter Abbeel ([email protected]).
from util import manhattanDistance
from game import Directions
import random, util
from game import Agent
class ReflexAgent(Agent):
"""
A reflex agent chooses an action at each choice point by examining
its alternatives via a state evaluation function.
The code below is provided as a guide. You are welcome to change
it in any way you see fit, so long as you don't touch our method
headers.
"""
def getAction(self, gameState):
"""
You do not need to change this method, but you're welcome to.
getAction chooses among the best options according to the evaluation function.
Just like in the previous project, getAction takes a GameState and returns
some Directions.X for some X in the set {North, South, West, East, Stop}
"""
# Collect legal moves and successor states
legalMoves = gameState.getLegalActions()
# Choose one of the best actions
scores = [self.evaluationFunction(gameState, action) for action in legalMoves]
bestScore = max(scores)
bestIndices = [index for index in range(len(scores)) if scores[index] == bestScore]
chosenIndex = random.choice(bestIndices) # Pick randomly among the best
"Add more of your code here if you want to"
return legalMoves[chosenIndex]
def evaluationFunction(self, currentGameState, action):
"""
Design a better evaluation function here.
The evaluation function takes in the current and proposed successor
GameStates (pacman.py) and returns a number, where higher numbers are better.
The code below extracts some useful information from the state, like the
remaining food (newFood) and Pacman position after moving (newPos).
newScaredTimes holds the number of moves that each ghost will remain
scared because of Pacman having eaten a power pellet.
Print out these variables to see what you're getting, then combine them
to create a masterful evaluation function.
"""
# Useful information you can extract from a GameState (pacman.py)
successorGameState = currentGameState.generatePacmanSuccessor(action)
newPos = successorGameState.getPacmanPosition()
newFood = successorGameState.getFood()
newGhostStates = successorGameState.getGhostStates()
newScaredTimes = [ghostState.scaredTimer for ghostState in newGhostStates]
"*** YOUR CODE HERE ***"
newScore = successorGameState.getScore()
import util
newFoodList = newFood.asList()
# Get the closest distance to a food pellet.
newFoodDistances = [util.manhattanDistance(foodPos, newPos) for foodPos in newFoodList]
minDistance = min(newFoodDistances) if len(newFoodDistances) > 0 else None
# Get the closest distance to any ghost (won't remember which gost if more than one ghost)
newGhostPositions = [ghostState.getPosition() for ghostState in newGhostStates]
newGhostDistances = [util.manhattanDistance(ghostPos, newPos) for ghostPos in newGhostPositions]
minGhostDistance = min(newGhostDistances) if len(newGhostDistances) > 0 else None
# Get closet distance to a capsule (i.e. food-pellet make ghosts scared)
capsuleList = currentGameState.getCapsules()
newCapsuleList = successorGameState.getCapsules()
capsuleDistances = [util.manhattanDistance(pos, newPos) for pos in newCapsuleList]
minCapsuleDistance = min(capsuleDistances) if len(capsuleDistances) > 0 else None
if len(capsuleList) > len(newCapsuleList):
newScore += 1
# Get the minimum amount of time that any ghost is still scared (just to be cautious).
minScaredTime = min(newScaredTimes)
# Scoring
minDistanceScore = 1.0 / float(minDistance + 1) if minDistance != None else 0
ghostDistanceScore = 1.0 / float(minGhostDistance + 1) if minGhostDistance != None else 0
minCapsuleDistanceScore = 1.0 / float(minCapsuleDistance + 1) if minCapsuleDistance != None else 0
if (minGhostDistance != None) and (minGhostDistance <= 1):
# If any ghost is normal, run from them if one of them is really close.
# Only if all of them scared, go to them.
return -1 if minScaredTime == 0 else newScore + 1
else:
# If any ghost is normal, get the closest food-pellet or capsule
# IF all ghost scared, go to them.
score = max(minCapsuleDistanceScore, minDistanceScore) if minScaredTime == 0 else ghostDistanceScore
return newScore + score
def scoreEvaluationFunction(currentGameState):
"""
This default evaluation function just returns the score of the state.
The score is the same one displayed in the Pacman GUI.
This evaluation function is meant for use with adversarial search agents
(not reflex agents).
"""
return currentGameState.getScore()
class MultiAgentSearchAgent(Agent):
"""
This class provides some common elements to all of your
multi-agent searchers. Any methods defined here will be available
to the MinimaxPacmanAgent, AlphaBetaPacmanAgent & ExpectimaxPacmanAgent.
You *do not* need to make any changes here, but you can if you want to
add functionality to all your adversarial search agents. Please do not
remove anything, however.
Note: this is an abstract class: one that should not be instantiated. It's
only partially specified, and designed to be extended. Agent (game.py)
is another abstract class.
"""
def __init__(self, evalFn = 'scoreEvaluationFunction', depth = '2'):
self.index = 0 # Pacman is always agent index 0
self.evaluationFunction = util.lookup(evalFn, globals())
self.depth = int(depth)
class MinimaxAgent(MultiAgentSearchAgent):
"""
Your minimax agent (question 2)
"""
def getAction(self, gameState):
"""
Returns the minimax action from the current gameState using self.depth
and self.evaluationFunction.
Here are some method calls that might be useful when implementing minimax.
gameState.getLegalActions(agentIndex):
Returns a list of legal actions for an agent
agentIndex=0 means Pacman, ghosts are >= 1
gameState.generateSuccessor(agentIndex, action):
Returns the successor game state after an agent takes an action
gameState.getNumAgents():
Returns the total number of agents in the game
gameState.isWin():
Returns whether or not the game state is a winning state
gameState.isLose():
Returns whether or not the game state is a losing state
"""
"*** YOUR CODE HERE ***"
bestAction = minimax(self, gameState, self.index, self.depth, None)
return bestAction[1][0]
def minimax(agent, gameState, agentIndex, ply, alphaBeta):
import sys
isMaximizingPlayer = (agentIndex == 0)
if (isMaximizingPlayer and ply == 0) or gameState.isWin() or gameState.isLose():
return agent.evaluationFunction(gameState), []
actions = gameState.getLegalActions(agentIndex)
if isMaximizingPlayer:
return max_value(agent, gameState, agentIndex, ply, actions, alphaBeta)
else:
return min_value(agent, gameState, agentIndex, ply, actions, alphaBeta)
def max_value(agent, gameState, agentIndex, ply, actions, alphaBeta):
import sys
nextAgentIndex = (agentIndex + 1) % gameState.getNumAgents()
# Save and restore the alphaBeta so that it only propagates from path to root, not over the entire tree.
savedAlphaBeta = [] + alphaBeta if alphaBeta != None else None
bestValue = -(sys.maxint-1), []
for action in actions:
successorState = gameState.generateSuccessor(agentIndex, action)
minimaxResult = minimax(agent, successorState, nextAgentIndex, ply - 1, alphaBeta)
value = minimaxResult[0], [action] + minimaxResult[1]
if value[0] > bestValue[0]:
bestValue = value
if alphaBeta != None:
if (bestValue[0] > alphaBeta[1]):
alphaBeta[0] = savedAlphaBeta[0]
alphaBeta[1] = savedAlphaBeta[1]
return bestValue
alphaBeta[0] = max(alphaBeta[0], bestValue[0])
if savedAlphaBeta != None:
alphaBeta[0] = savedAlphaBeta[0]
alphaBeta[1] = savedAlphaBeta[1]
return bestValue
def min_value(agent, gameState, agentIndex, ply, actions, alphaBeta):
import sys
nextAgentIndex = (agentIndex + 1) % gameState.getNumAgents()
# Save and restore the alphaBeta so that it only propagates from path to root, not over the entire tree.
savedAlphaBeta = [] + alphaBeta if alphaBeta != None else None
bestValue = sys.maxint, []
for action in actions:
successorState = gameState.generateSuccessor(agentIndex, action)
minimaxResult = minimax(agent, successorState, nextAgentIndex, ply, alphaBeta)
value = minimaxResult[0], [action] + minimaxResult[1]
if value[0] < bestValue[0]:
bestValue = value
if alphaBeta != None:
if (bestValue[0] < alphaBeta[0]):
alphaBeta[0] = savedAlphaBeta[0]
alphaBeta[1] = savedAlphaBeta[1]
return bestValue
alphaBeta[1] = min(alphaBeta[1], bestValue[0])
if savedAlphaBeta != None:
alphaBeta[0] = savedAlphaBeta[0]
alphaBeta[1] = savedAlphaBeta[1]
return bestValue
class AlphaBetaAgent(MultiAgentSearchAgent):
"""
Your minimax agent with alpha-beta pruning (question 3)
"""
def getAction(self, gameState):
"""
Returns the minimax action using self.depth and self.evaluationFunction
"""
"*** YOUR CODE HERE ***"
import sys
alphaBeta = [-(sys.maxint-1), sys.maxint]
bestAction = minimax(self, gameState, self.index, self.depth, alphaBeta)
return bestAction[1][0]
class ExpectimaxAgent(MultiAgentSearchAgent):
"""
Your expectimax agent (question 4)
"""
def getAction(self, gameState):
"""
Returns the expectimax action using self.depth and self.evaluationFunction
All ghosts should be modeled as choosing uniformly at random from their
legal moves.
"""
"*** YOUR CODE HERE ***"
bestAction = expectimax(self, gameState, self.index, self.depth)
return bestAction[1][0]
def betterEvaluationFunction(gameState):
# See inline comments for explanations.
newPos = gameState.getPacmanPosition()
newFood = gameState.getFood()
newGhostStates = gameState.getGhostStates()
newScaredTimes = [ghostState.scaredTimer for ghostState in newGhostStates]
newScore = gameState.getScore()
# maximum manhattan distance to any object/agent.
maxDistance = gameState.data.layout.height + gameState.data.layout.width
import util
newFoodList = newFood.asList()
# Get the closest distance to a food pellet.
newFoodDistances = [util.manhattanDistance(foodPos, newPos) for foodPos in newFoodList]
minDistance = min(newFoodDistances) if len(newFoodDistances) > 0 else None
# Get average distance to a a food pellet, i.e. the center of all food-pellets.
if len(newFoodList) > 0:
totalDistance = reduce(lambda pos1, pos2: (float(pos1[0]+pos2[0]), float(pos1[1]+pos2[1])), newFoodList)
averageDistance = util.manhattanDistance((totalDistance[0] / float(len(newFoodList)), totalDistance[1] / float(len(newFoodList))), newPos)
else:
averageDistance = None
# Get the closest distance to any ghost (won't remember which ghost if more than one ghost agent in game)
newGhostPositions = [ghostState.getPosition() for ghostState in newGhostStates]
newGhostDistances = [util.manhattanDistance(ghostPos, newPos) for ghostPos in newGhostPositions]
minGhostDistance = min(newGhostDistances)
# Get closet distance to a capsule (i.e. food-pellet that makes ghosts scared)
newCapsuleList = gameState.getCapsules()
capsuleDistances = [util.manhattanDistance(pos, newPos) for pos in newCapsuleList]
minCapsuleDistance = min(capsuleDistances) if len(capsuleDistances) > 0 else None
# Any eaten capsule is worth 20 points for scoring.
# This tends to increase scared-times of ghosts, increasing possible overall average scores for pacman.
# It may decrease possible maximum scores, though.
if len(newCapsuleList) == 0:
newScore += 20
# Get the minimum amount of time that any ghost is still scared (just to be cautious).
minScaredTime = min(newScaredTimes)
# Scoring
minDistanceScore = 1.0 / float(minDistance + 1) if minDistance != None else 0
averageDistanceScore = 1.0 / float(averageDistance + 1) if averageDistance != None else 0
ghostDistanceScore = 1.0 / float(minGhostDistance + 1)
minCapsuleDistanceScore = 1.0 / float(minCapsuleDistance + 1) if minCapsuleDistance != None else 0
if (minGhostDistance <= 1):
# If any ghost is normal, run from them if one of them is really close.
# Only if all of them are scared, go to them (be cautious).
return -10 if minScaredTime == 0 else newScore + 10
else:
# How hungry is pacman? Go to closest food-pellet. When all things equal, gravitate towards
# center of all food-pellets.
hungryScore = 0.1 * averageDistanceScore + max(minCapsuleDistanceScore, minDistanceScore)
# How afraid is pacman? The further the distance from the closest ghost, the better.
# Weight of this score is less than the hungry score.
afraidScore = 0.9 * (minGhostDistance / maxDistance)
# If any ghost is normal, get the closest food-pellet or capsule
# Only if all ghosts are scared, go to them:
score = (hungryScore + afraidScore) if minScaredTime == 0 else (10.0 * ghostDistanceScore)
return newScore + score
def expectimax(agent, gameState, agentIndex, ply):
import sys
isMaximizingPlayer = (agentIndex == 0)
if (isMaximizingPlayer and ply == 0) or gameState.isWin() or gameState.isLose():
return agent.evaluationFunction(gameState), []
actions = gameState.getLegalActions(agentIndex)
if isMaximizingPlayer:
return max_value2(agent, gameState, agentIndex, ply, actions)
else:
return expected_value(agent, gameState, agentIndex, ply, actions)
def max_value2(agent, gameState, agentIndex, ply, actions):
import sys
nextAgentIndex = (agentIndex + 1) % gameState.getNumAgents()
bestValue = -(sys.maxint-1), []
for action in actions:
successorState = gameState.generateSuccessor(agentIndex, action)
minimaxResult = expectimax(agent, successorState, nextAgentIndex, ply - 1)
value = minimaxResult[0], [action] + minimaxResult[1]
if value[0] > bestValue[0]:
bestValue = value
return bestValue
def expected_value(agent, gameState, agentIndex, ply, actions):
nextAgentIndex = (agentIndex + 1) % gameState.getNumAgents()
bestValue = 0, []
chance = 1.0 / float(len(actions))
for action in actions:
successorState = gameState.generateSuccessor(agentIndex, action)
minimaxResult = expectimax(agent, successorState, nextAgentIndex, ply)
expectedValue = minimaxResult[0]
bestValue = bestValue[0] + chance * expectedValue, minimaxResult[1]
return bestValue
# Abbreviation
better = betterEvaluationFunction