/
multiAgents.py
595 lines (474 loc) · 21.4 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
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
# multiAgents.py
# --------------
# Licensing Information: Please do not distribute or publish solutions to this
# project. You are free to use and extend these projects for educational
# purposes. The Pacman AI projects were developed at UC Berkeley, primarily by
# John DeNero (denero@cs.berkeley.edu) and Dan Klein (klein@cs.berkeley.edu).
# Student side autograding was added by Brad Miller, Nick Hay, and Pieter
# Abbeel in Spring 2013.
# For more info, see http://inst.eecs.berkeley.edu/~cs188/pacman/pacman.html
import random
import util
from game import Agent, Directions
MAX_VALUE = float("inf")
MIN_VALUE = float("-inf")
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
legal_moves = gameState.getLegalActions()
# Choose one of the best actions
scores = [self.evaluationFunction(gameState, action) for action in legal_moves]
best_score = max(scores)
best_indices = [index for index in range(len(scores)) if scores[index] == best_score]
chosen_index = random.choice(best_indices) # Pick randomly among the best
"Add more of your code here if you want to"
return legal_moves[chosen_index]
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)
successor_game_state = currentGameState.generatePacmanSuccessor(action)
new_pos = successor_game_state.getPacmanPosition()
new_food = successor_game_state.getFood()
new_ghost = successor_game_state.getGhostStates()
walls = successor_game_state.getWalls()
ghost_pos = successor_game_state.getGhostPositions()
"*** YOUR CODE HERE ***"
# If this is win state, return the maximum value
if successor_game_state.isWin():
return MAX_VALUE
# Hold negative position on stop, encourage agent to take other actions
if action == Directions.STOP:
next_score = 0
else:
next_score = 1
# Calculate the ghost related score
for i in range(len(ghost_pos)):
distance = util.manhattanDistance(new_pos, ghost_pos[i])
# which means ghost is eatable
if new_ghost[i].scaredTimer > 0:
next_score += new_ghost[i].scaredTimer - distance
else:
# This position is very dangerous, so use the minimal value to represent
if distance < 2:
return MIN_VALUE
# encourage the agent keep away from ghost, regard distance larger than 5 as safe
elif distance > 5:
next_score += 5
else:
next_score += distance
# If new position has a food, add 10 score
if currentGameState.getFood()[new_pos[0]][new_pos[1]]:
next_score += 10
# Else use bfs to find the nearest food position
else:
pos_to_explore = []
explored_pos = set()
pos_to_explore.append((new_pos, 0))
distance = 0
while pos_to_explore:
frontier = pos_to_explore.pop(0)
explored_pos.add(frontier[0])
if new_food[frontier[0][0]][frontier[0][1]]:
distance = frontier[1]
break
else:
successor_pos = [(frontier[0][0] + 1, frontier[0][1]), (frontier[0][0] - 1, frontier[0][1]),
(frontier[0][0], frontier[0][1] + 1), (frontier[0][0], frontier[0][1] - 1)]
for pos in successor_pos:
if pos[0] < 0 or pos[0] >= new_food.width or pos[1] < 0 or pos[1] >= new_food.height:
continue
if not walls[pos[0]][pos[1]] and pos not in explored_pos:
pos_to_explore.append((pos, frontier[1] + 1))
next_score -= distance
return next_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'):
Agent.__init__(self)
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 minmax 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
"""
"*** YOUR CODE HERE ***"
# init some parameter for next steps
possible_next_action_list = gameState.getLegalActions(0)
max_value = MIN_VALUE
next_action = Directions.STOP
agent_num = gameState.getNumAgents()
# get next agent index and next depth
if agent_num > 1:
next_agent = self.index + 1
depth = 0
else:
next_agent = self.index
depth = 1
# Travel all the possible actions to determine the best action
for action in possible_next_action_list:
next_state = gameState.generateSuccessor(0, action)
if next_state.isWin():
return action
current_value = self.evaluate_action(next_agent, depth, next_state)
if current_value > max_value:
max_value = current_value
next_action = action
return next_action
def evaluate_action(self, agent_index, depth, game_state):
'''
Evaluation function to calculate the action score
:param agent_index: If agent index is 0, act as max_value, other wise min_value
:param depth: current depth
:param game_state: current game state
:return: the score of current action
'''
# Get the agent number
agent_num = game_state.getNumAgents()
# Check whether both Pacman or ghosts moving enough times or current state is a goal state
if depth == self.depth or game_state.isWin() or game_state.isLose():
return self.evaluationFunction(game_state)
# If all ghosts moved, next moved actor would be Pacman
elif agent_index == agent_num - 1:
depth += 1
next_agent_index = self.index
# Get next ghost agent index
else:
next_agent_index = agent_index + 1
# Init some values
possible_next_action_list = game_state.getLegalActions(agent_index)
min_value = MAX_VALUE
max_value = MIN_VALUE
for action in possible_next_action_list:
next_state = game_state.generateSuccessor(agent_index, action)
# if next_state is goal state, then there is no need to further evaluate
if next_state.isWin() or next_state.isLose():
current_value = self.evaluationFunction(next_state)
# Recursively calculate the state cost
else:
current_value = self.evaluate_action(next_agent_index, depth, next_state)
# Update the max and min value
if current_value > max_value:
max_value = current_value
if current_value < min_value:
min_value = current_value
# If agent is Pacman, then it need the max_value, otherwise, it will need the min_value
if agent_index == self.index:
return max_value
else:
return min_value
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 ***"
# init some parameter for next steps
possible_next_action_list = gameState.getLegalActions(0)
max_value = MIN_VALUE
next_action = Directions.STOP
agent_num = gameState.getNumAgents()
# get next agent index and next depth
if agent_num > 1:
next_agent = self.index + 1
depth = 0
else:
next_agent = self.index
depth = 1
# Travel all the possible actions to determine the best action
alpha = MIN_VALUE
beta = MAX_VALUE
for action in possible_next_action_list:
next_state = gameState.generateSuccessor(0, action)
if next_state.isWin():
return action
current_value = self.evaluate_action(next_agent, depth, next_state, alpha, beta)
if current_value > max_value:
max_value = current_value
next_action = action
alpha = max(alpha, max_value)
return next_action
def evaluate_action(self, agent_index, depth, game_state, alpha, beta):
'''
Evaluate the action performance
:param agent_index: the index of action that need to be evaluated
:param depth: how many depth has been explored
:param game_state: Current game state
:param alpha: Current alpha
:param beta: Current beta
:return: The score of current action
'''
# Check whether both Pacman or ghosts moving enough times or current state is a goal state
if depth == self.depth or game_state.isWin() or game_state.isLose():
return self.evaluationFunction(game_state)
# Based on the agent type, decide which function should be used
if agent_index == 0:
return self.max_value(game_state, depth, alpha, beta)
else:
return self.min_value(game_state, depth, agent_index, alpha, beta)
def max_value(self, game_state, depth, alpha, beta):
'''
Function to evaluate the max agent
:param game_state: current game state
:param depth: current game depth
:param alpha: current alpha
:param beta: current beta
:return: current score
'''
# Get the agent number
agent_num = game_state.getNumAgents()
# We can make sure that this state is not the goal state, so Just generate successors directly
possible_action_list = game_state.getLegalActions(0)
max_value = MIN_VALUE
# Determine next agent and depth index
if agent_num == 1:
next_agent = 0
next_depth = depth + 1
else:
next_agent = 1
next_depth = depth
# Travel all possible states
for action in possible_action_list:
next_game_state = game_state.generateSuccessor(0, action)
if next_game_state.isWin() or next_game_state.isLose():
current_value = self.evaluationFunction(next_game_state)
else:
current_value = self.evaluate_action(next_agent, next_depth, next_game_state, alpha, beta)
max_value = max(current_value, max_value)
# If max value greater than beta, we can prune the following path
if max_value > beta:
return max_value
# Get new alpha
alpha = max(alpha, max_value)
return max_value
def min_value(self, game_state, depth, agent_index, alpha, beta):
'''
Function to evaluate the max agent
:param game_state: current game state
:param depth: current game depth
:param alpha: current alpha
:param beta: current beta
:param agent_index: The agent that need to be judged
:return: current score
'''
# Get the agent number
agent_num = game_state.getNumAgents()
# We can make sure that this state is not the goal state, so Just generate successors directly
possible_action_list = game_state.getLegalActions(agent_index)
min_value = MAX_VALUE
# Determine next agent and depth index
if agent_num == agent_index + 1:
next_agent = 0
next_depth = depth + 1
else:
next_agent = agent_index + 1
next_depth = depth
# Travel all possible states
for action in possible_action_list:
next_game_state = game_state.generateSuccessor(agent_index, action)
if next_game_state.isWin() or next_game_state.isLose():
current_value = self.evaluationFunction(next_game_state)
else:
current_value = self.evaluate_action(next_agent, next_depth, next_game_state, alpha, beta)
min_value = min(current_value, min_value)
# If max value greater than alpha, we can prune the following path
if min_value < alpha:
return min_value
# get new beta
beta = min(beta, min_value)
return min_value
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 ***"
# init some parameter for next steps
max_value = MIN_VALUE
next_action = Directions.STOP
if gameState.isLose() or gameState.isWin():
return next_action
possible_next_action_list = gameState.getLegalActions(0)
agent_num = gameState.getNumAgents()
# get next agent index and next depth
if agent_num > 1:
next_agent = self.index + 1
depth = 0
else:
next_agent = self.index
depth = 1
# Travel all the possible actions to determine the best action
for action in possible_next_action_list:
next_state = gameState.generateSuccessor(0, action)
if next_state.isWin():
return action
current_value = self.evaluate_action(next_agent, depth, next_state)
if current_value > max_value:
max_value = current_value
next_action = action
return next_action
def evaluate_action(self, agent_index, depth, game_state):
'''
Evaluation function to calculate the action score
'''
# Get the agent number
agent_num = game_state.getNumAgents()
# Check whether both Pacman or ghosts moving enough times or current state is a goal state
if depth == self.depth or game_state.isWin() or game_state.isLose():
return self.evaluationFunction(game_state)
# If all ghosts moved, next moved actor would be Pacman
elif agent_index == agent_num - 1:
depth += 1
next_agent_index = self.index
# Get next ghost agent index
else:
next_agent_index = agent_index + 1
# Init some values
possible_next_action_list = game_state.getLegalActions(agent_index)
max_value = MIN_VALUE
total_value = 0
for action in possible_next_action_list:
next_state = game_state.generateSuccessor(agent_index, action)
# if next_state is goal state, then there is no need to further evaluate
if next_state.isWin() or next_state.isLose():
current_value = self.evaluationFunction(next_state)
# Recursively calculate the state cost
else:
current_value = self.evaluate_action(next_agent_index, depth, next_state)
# Update the max and total value
if current_value > max_value:
max_value = current_value
total_value += current_value
# If agent is Pacman, then it need the max_value, otherwise, it will need the average value
if agent_index == self.index:
return max_value
else:
return total_value / float(len(possible_next_action_list))
def betterEvaluationFunction(currentGameState):
"""
Your extreme ghost-hunting, pellet-nabbing, food-gobbling, unstoppable
evaluation function (question 5).
DESCRIPTION: <write something here so we know what you did>
"""
"*** YOUR CODE HERE ***"
if currentGameState.isWin() or currentGameState.isLose():
return currentGameState.getScore()
pacman_pos = currentGameState.getPacmanPosition()
ghost_pos = currentGameState.getGhostPositions()
ghost_state = currentGameState.getGhostStates()
current_value = currentGameState.getScore()
food_pos = currentGameState.getFood()
wall_pos = currentGameState.getWalls()
min_distance = MAX_VALUE
# Calculate the ghost related score
for i in range(len(ghost_pos)):
distance = util.manhattanDistance(pacman_pos, ghost_pos[i])
# which means ghost is eatable
if ghost_state[i].scaredTimer > 0:
current_value += (100 / distance) if ghost_state[i].scaredTimer > distance else 0
elif distance < min_distance:
min_distance = distance
# Based on min distance
if min_distance < 2:
return current_value - 500
else:
current_value -= 50 / max(min_distance, 5)
# Use BFS to find the nearest food position
pos_to_explore = []
explored_pos = set()
pos_to_explore.append((pacman_pos, 0))
distance = 0
while pos_to_explore:
frontier = pos_to_explore.pop(0)
explored_pos.add(frontier[0])
if food_pos[frontier[0][0]][frontier[0][1]]:
distance = frontier[1]
break
else:
successor_pos = [(frontier[0][0] + 1, frontier[0][1]), (frontier[0][0] - 1, frontier[0][1]),
(frontier[0][0], frontier[0][1] + 1), (frontier[0][0], frontier[0][1] - 1)]
for pos in successor_pos:
if pos[0] < 0 or pos[0] >= food_pos.width or pos[1] < 0 or pos[1] >= food_pos.height:
continue
if not wall_pos[pos[0]][pos[1]] and pos not in explored_pos:
pos_to_explore.append((pos, frontier[1] + 1))
# make sure that the nearest get the highest score and will not affect if current point has a food
current_value += 6.0 / distance
return current_value
# Abbreviation
better = betterEvaluationFunction
class ContestAgent(MultiAgentSearchAgent):
"""
Your agent for the mini-contest
"""
def getAction(self, gameState):
"""
Returns an action. You can use any method you want and search to any depth you want.
Just remember that the mini-contest is timed, so you have to trade off speed and computation.
Ghosts don't behave randomly anymore, but they aren't perfect either -- they'll usually
just make a beeline straight towards Pacman (or away from him if they're scared!)
"""
"*** YOUR CODE HERE ***"
util.raiseNotDefined()