-
Notifications
You must be signed in to change notification settings - Fork 0
/
FTD.py
295 lines (249 loc) · 10.1 KB
/
FTD.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
# baselineTeam.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).
# For more info, see http://inst.eecs.berkeley.edu/~cs188/sp09/pacman.html
from captureAgents import CaptureAgent
import distanceCalculator
import random, time, util
from game import Directions
import game
from util import nearestPoint
#################
# Team creation #
#################
def createTeam(firstIndex, secondIndex, isRed, first = 'OffensiveReflexAgent', second = 'DefensiveReflexAgent'):
"""
This function should return a list of two agents that will form the
team, initialized using firstIndex and secondIndex as their agent
index numbers. isRed is True if the red team is being created, and
will be False if the blue team is being created.
As a potentially helpful development aid, this function can take
additional string-valued keyword arguments ("first" and "second" are
such arguments in the case of this function), which will come from
the --redOpts and --blueOpts command-line arguments to capture.py.
For the nightly contest, however, your team will be created without
any extra arguments, so you should make sure that the default
behavior is what you want for the nightly contest.
"""
return [eval(first)(firstIndex), eval(second)(secondIndex)]
##########
# Agents #
##########
class ReflexCaptureAgent(CaptureAgent):
"""
A base class for reflex agents that chooses score-maximizing actions
"""
def chooseAction(self, gameState):
"""
Picks among the actions with the highest Q(s,a).
"""
actions = gameState.getLegalActions(self.index)
# You can profile your evaluation time by uncommenting these lines
# start = time.time()
values = [self.evaluate(gameState, a) for a in actions]
# print 'eval time for agent %d: %.4f' % (self.index, time.time() - start)
maxValue = max(values)
bestActions = [a for a, v in zip(actions, values) if v == maxValue]
return random.choice(bestActions)
def getSuccessor(self, gameState, action):
"""
Finds the next successor which is a grid position (location tuple).
"""
successor = gameState.generateSuccessor(self.index, action)
pos = successor.getAgentState(self.index).getPosition()
if pos != nearestPoint(pos):
# Only half a grid position was covered
return successor.generateSuccessor(self.index, action)
else:
return successor
def evaluate(self, gameState, action):
"""
Computes a linear combination of features and feature weights
"""
features = self.getFeatures(gameState, action)
weights = self.getWeights(gameState, action)
return features * weights
def getFeatures(self, gameState, action):
"""
Returns a counter of features for the state
"""
features = util.Counter()
successor = self.getSuccessor(gameState, action)
features['successorScore'] = self.getScore(successor)
return features
def getWeights(self, gameState, action):
"""
Normally, weights do not depend on the gamestate. They can be either
a counter or a dictionary.
"""
return {'successorScore': 1.0}
class OffensiveReflexAgent(CaptureAgent):
"""
A reflex agent that seeks food. This is an agent
we give you to get an idea of what an offensive agent might look like,
but it is by no means the best or only way to build an offensive agent.
"""
def __init__( self, index, timeForComputing = .1 ):
CaptureAgent.__init__(self, index, timeForComputing)
self.discount = 0.9
self.noise = 0
self.alpha = 0.01
self.epsilon = 0.05
self.preScore = 0
self.preState = None
self.preAction = None
self.weights = util.Counter()
self.weights = util.Counter()
self.weights['distanceToFood'] = -1
self.weights['successorScore'] = 100
#self.weights['onOffence'] = 0
#self.weights['numDefenders'] = 0
#self.weights['defenderDistance'] = 0
#self.weights['attackerDistance'] = 0
self.weights['foodsLeft'] = -1
self.weights['foodsRemained'] = 1
self.weights = util.normalize(self.weights)
def chooseAction(self, gameState):
"""
Picks among the actions with the highest Q(s,a).
"""
score = len(self.getFood(gameState).asList())
reward = (score - self.preScore) * 50
self.update(reward, gameState)
actions = gameState.getLegalActions(self.index)
if util.flipCoin(self.epsilon): #exploration
action = random.choice(actions)
else:
action = self.getPolicy(gameState, actions)
# You can profile your evaluation time by uncommenting these lines
# start = time.time()
# print 'eval time for agent %d: %.4f' % (self.index, time.time() - start)
self.preState = gameState
self.preAction = action
self.preScore = score
return action
def getPolicy(self, gameState, actions):
if len(actions) == 0:
return None
q_values = []
for action in actions:
q_values.append(self.getQValue(gameState, action))
index = q_values.index(max(q_values))
return actions[index]
def getSuccessor(self, gameState, action):
"""
Finds the next successor which is a grid position (location tuple).
"""
successor = gameState.generateSuccessor(self.index, action)
pos = successor.getAgentState(self.index).getPosition()
if pos != nearestPoint(pos):
# Only half a grid position was covered
return successor.generateSuccessor(self.index, action)
else:
return successor
def getFeatures(self, gameState, action):
features = util.Counter()
successor = self.getSuccessor(gameState, action)
myState = successor.getAgentState(self.index)
myPos = myState.getPosition()
features['successorScore'] = self.getScore(successor)
features['onOffence'] = 0
if myState.isPacman: features['onOffence'] = 1
# Compute distance to the nearest food
foodList = self.getFood(successor).asList()
myPos = successor.getAgentState(self.index).getPosition()
features['foodsLeft'] = len(foodList)
minDistance = min([self.getMazeDistance(myPos, food) for food in foodList])
features['distanceToFood'] = minDistance
friends = [successor.getAgentState(i) for i in self.getTeam(successor)]
enemies = [successor.getAgentState(i) for i in self.getOpponents(successor)]
attackers = [a for a in friends if a.isPacman and a.getPosition() != None]
defenders = [a for a in enemies if not a.isPacman and a.getPosition() != None]
#features['numDefenders'] = len(defenders)
if len(attackers) > 0:
dists = [self.getMazeDistance(myPos, a.getPosition()) for a in attackers]
#features['attackerDistance'] = max(dists)
if len(defenders) > 0:
dists = [self.getMazeDistance(myPos, a.getPosition()) for a in defenders]
#features['defenderDistance'] = min(dists)
return features
def getWeights(self, gameState, action):
return self.weights
def update(self, reward, gameState):
if self.preState == None:
return
if not gameState.getAgentState(self.index).isPacman: #don't update unless it is in offense position
return
correction = (reward + self.discount * self.getValue(gameState)) - self.getQValue(self.preState, self.preAction)
#print correction
features = self.getFeatures(self.preState, self.preAction)
for feature in features:
#print feature
#print self.weights[feature]
self.weights[feature] = self.weights[feature] + self.alpha * correction * features[feature]
self.weights = util.normalize(self.weights)
print self.weights
def getValue(self, gameState):
"""
Returns max_action Q(state,action)
where the max is over legal actions. Note that if
there are no legal actions, which is the case at the
terminal state, you should return a value of 0.0.
"""
actions = gameState.getLegalActions(self.index)
if len(actions) == 0:
return 0
q_values = []
for action in actions:
q_values.append(self.getQValue(gameState, action))
return max(q_values)
def getQValue(self, gameState, action):
"""
Should return Q(state,action) = w * featureVector
where * is the dotProduct operator
"""
qValue = 0
features = self.getFeatures(gameState, action)
for feature in features:
qValue += self.weights[feature] * features[feature]
return qValue
class DefensiveReflexAgent(ReflexCaptureAgent):
"""
A reflex agent that keeps its side Pacman-free. Again,
this is to give you an idea of what a defensive agent
could be like. It is not the best or only way to make
such an agent.
"""
def __init__( self, index, timeForComputing = .1 ):
ReflexCaptureAgent.__init__(self, index, timeForComputing)
self.weights = util.Counter()
self.weights['numInvaders'] = -1000
self.weights['onDefense'] = 100
self.weights['invaderDistance'] = -10
self.weights['stop'] = -100
self.weights['reverse'] = -2
#self.weights = util.normalize(self.weights)
def getFeatures(self, gameState, action):
features = util.Counter()
successor = self.getSuccessor(gameState, action)
myState = successor.getAgentState(self.index)
myPos = myState.getPosition()
# Computes whether we're on defense (1) or offense (0)
features['onDefense'] = 1
if myState.isPacman: features['onDefense'] = 0
# Computes distance to invaders we can see
enemies = [successor.getAgentState(i) for i in self.getOpponents(successor)]
invaders = [a for a in enemies if a.isPacman and a.getPosition() != None]
features['numInvaders'] = len(invaders)
if len(invaders) > 0:
dists = [self.getMazeDistance(myPos, a.getPosition()) for a in invaders]
features['invaderDistance'] = min(dists)
if action == Directions.STOP: features['stop'] = 1
rev = Directions.REVERSE[gameState.getAgentState(self.index).configuration.direction]
if action == rev: features['reverse'] = 1
return features
def getWeights(self, gameState, action):
return self.weights