def __init__(self, dsl=None):
if dsl is None:
dsl = DSL()
self._dsl = DSL()
self._non_terminal_symbols = self._dsl._grammar.keys()
self._rules = set()
self.generate_rules()
def get_action(self, state):
actions = state.available_moves()
for a in actions:
if DSL.hasWonColumn(state) and DSL.isStopAction(a):
self._counter_calls[0] += 1
return a
return actions[0]
def get_action(self, state):
actions = state.available_moves()
print("Actions: ", actions)
for a in actions:
if DSL.actionWinsColumn(state, a):
return a
if DSL.isDoubles(a):
return a
if DSL.containsNumber(a, (6,2)):
return a
if DSL.isStopAction(a):
return a
return actions[0]
def ESZ():
# parameters
populationSize = 30
numOfGene = 30
numOfTour = 5
numOfElite = 10
numOfRounds = 10
mutateRate = 0.5
# init population
dslExample = DSL()
scripts = dslExample.generateRandomScript(populationSize)
for i in scripts:
i.saveFile('')
scriptIdCounter = populationSize # show the id of the next script
numOfWinRule = []
# evaluate
for _ in range(numOfGene):
numOfWinRule.append(count_winning_rule(scripts))
print(count_winning_rule(scripts))
Evaluate(scripts, numOfRounds) # change the fitness numbers of scripts
# get elites
nextGene = elite(scripts, numOfElite)
# clear fitness and call count
for i in scripts:
i.clearAttributes()
while len(nextGene) < len(scripts):
p1, p2 = tournament(scripts, numOfTour, numOfRounds)
c = crossoverBackup(p1, p2, scriptIdCounter)
while len(c.getRules()) == 0:
c = crossoverBackup(p1, p2, scriptIdCounter)
mutate(c, mutateRate, dslExample)
c.saveFile('')
scriptIdCounter += 1
nextGene.append(c)
scripts = deepcopy(nextGene)
# keep the best scripts
Evaluate(scripts, numOfRounds)
sortedScripts = sorted(scripts, key=lambda x: x.getFitness(), reverse=True)
winnerPath = 'winner/'
if not os.path.exists(os.path.split(winnerPath)[0]):
os.makedirs(os.path.split(winnerPath)[0])
sortedScripts[0].saveFile(winnerPath)
for i in range(numOfGene):
print(
"# of scripts containing winning rule in generation " + str(i) +
": ", str(numOfWinRule[i]))
def get_action(self, state):
actions = state.available_moves()
if DSL.isYNAction(actions):
if DSL.hasAvailableNeuralMarker(state):
return 'y'
if DSL.hasWonColumn(state):
return 'n'
if DSL.continueBecausehighProbNotBust(state):
return 'y'
return random.choice(actions)
else:
if DSL.hasAvailableNeuralMarker(state):
actions = deepcopy(DSL.actionsUseLessMarker(state, actions))
for a in actions:
if DSL.isDoubles(a):
return a
for a in actions:
if DSL.actionWinsColumn(state, a):
return a
# should be later than detect the already progressed one (use of neutral marker)
# newActions = []
# if DSL.hasAvailableNeuralMarker(state):
# for a in actions:
# if DSL.actionEasyToPrograss(state, a):
# newActions.append(a)
# if newActions:
# actions = deepcopy(newActions)
return random.choice(actions)
def get_action(self, state):
actions = state.available_moves()
for a in actions:
if DSL.containsNumber(a,5):
self._counter_calls[0] += 1
return a
if DSL.isStopAction(a) and DSL.isStopAction(a):
self._counter_calls[1] += 1
return a
return actions[0]
def get_action(self, state):
actions = state.available_moves()
for a in actions:
if DSL.isDoubles(a):
return a
return actions[0]
def get_action(self, state):
actions = state.available_moves()
# print("Actions: ", actions)
for a in actions:
if DSL.isStopAction(a):
return a
return actions[0]
def train_rule(trainingData):
# init parameters
newDsl = DSL()
sampleNumber = 1000
sampledRule = []
prevRule = proposal_function_node(newDsl, 'B', 'S')
prevRule.expand()
prevScore = 0
# sample "sampleNumber" times, create less samples than this number
for i in range(sampleNumber):
# generate a new sample
# newRule.generatedRule is the newly sampled rule
newRule = deepcopy(prevRule)
error = 0
allNodes = []
allNodes += newRule.getAllChildNodes()
mutateNode = random.choice(allNodes)
mutateNode.mutate()
newRule.geneRule()
# evaluate possible new sample
never_used = True
for state_and_action in trainingData:
state = state_and_action[0]
chosen_action = state_and_action[1]
used = False
for a in state.available_moves():
if eval(newRule.generatedRule):
used = True
never_used = False
if a != chosen_action:
error += 1
break
if not used and state.available_moves()[0] != chosen_action:
error += 1
# decide sample or not
if never_used:
continue
score = exp(-0.5 * error)
if prevScore > 0 and score / prevScore < 1:
ran_num = random.uniform(0.0, 1.0)
if ran_num < score / prevScore:
sampledRule.append((newRule, error))
prevRule = deepcopy(newRule)
prevScore = score
else:
sampledRule.append((newRule, error))
prevRule = deepcopy(newRule)
prevScore = score
# best sampled rule is the rule with least errors
bestRule = min(sampledRule, key=lambda x: x[1])
return bestRule[0].generatedRule
def get_action(self, state):
actions = state.available_moves()
for a in actions:
if DSL.actionWinsColumn(state, a):
return a
if DSL.isDoubles(a):
return a
if DSL.containsNumber(a, 6) or DSL.containsNumber(a, 2):
if np.random.choice([0, 1], p=[0.6, 0.4]) > 0:
return a
if DSL.isStopAction(a):
# Return 'y' with 70% probability and 'n' with 30% probability.
if DSL.hasWonColumn(state, a):
return 'n'
else:
return 'y'
return actions[0]
def get_action(self, state):
actions = state.available_moves()
for a in actions:
if DSL.hasWonColumn(state) and DSL.isStopAction(a):
self._counter_calls[0] += 1
return a
if DSL.actionWinsColumn(state, a) and DSL.hasWonColumn(state):
self._counter_calls[1] += 1
return a
if DSL.numberPositionsConquered(
state, 5) > 2 and DSL.containsNumber(a, 4):
self._counter_calls[2] += 1
return a
return actions[0]
prev = prev.replace('B1', np.random.choice(dsl._grammar['B1']), 1)
self.prev = prev # This is the new rule.
# print("prev is ->", self.prev)
while 'SMALL_NUMBER' in prev:
prev = prev.replace('SMALL_NUMBER', np.random.choice(dsl._grammar['SMALL_NUMBER']), 1)
while 'NUMBER' in prev:
prev = prev.replace('NUMBER', np.random.choice(dsl._grammar['NUMBER']), 1)
updated_script = list()
updated_script.append(prev)
# print("updated_script is -> ", updated_script)
else:
new_script = self.prev
# print("previous script is ->", new_script)
while 'SMALL_NUMBER' in new_script:
new_script = new_script.replace('SMALL_NUMBER', np.random.choice(dsl._grammar['SMALL_NUMBER']), 1)
while 'NUMBER' in new_script:
new_script = new_script.replace('NUMBER', np.random.choice(dsl._grammar['NUMBER']),1)
updated_script = list()
updated_script.append(new_script)
# print("Newly updated script is -> ", updated_script)
return list(updated_script)
# The following snippet of code is for testing purposes!
# The tree representation of the rules seem to be working fine
dsl = DSL()
tree = Tree(dsl)
for i in range(3):
tree.change_tree()
def __init__(self):
self.dsl = DSL()
self.tree = Tree(self.dsl)
self.path = "C:\\Users\\Rohan\\Desktop\\cant-stop-assignment\\created_scripts\\"
def get_action(self, state):
actions = state.available_moves()
for a in actions:
if DSL.isDoubles(a) and DSL.actionWinsColumn(state,a):
self._counter_calls[0] += 1
return a
if DSL.numberPositionsConquered(state,5)>0 and DSL.containsNumber(a,4):
self._counter_calls[1] += 1
return a
if DSL.isStopAction(a) and DSL.numberPositionsConquered(state,5)>1 and DSL.containsNumber(a,6):
self._counter_calls[2] += 1
return a
if DSL.isStopAction(a) and DSL.numberPositionsProgressedThisRoundColumn(state,2)>2 and DSL.isStopAction(a):
self._counter_calls[3] += 1
return a
if DSL.isStopAction(a):
self._counter_calls[4] += 1
return a
if DSL.actionWinsColumn(state,a) and DSL.actionWinsColumn(state,a):
self._counter_calls[5] += 1
return a
if DSL.isStopAction(a):
self._counter_calls[6] += 1
return a
return actions[0]
def synthesize(self):
dsl = DSL()
# part3 = PlayerPart3()
indent_count = 1
script = dsl.start
line1 = script.replace(
'S', np.random.choice(dsl._grammar['S'], p=[0.9, 0.1]))
# line1 = line1 + ":"
# print(line1)
if line1 in ['']:
line1 = False
# print(line1)
return [[' DSL.isDoubles(a) ']]
while ('S' in line1):
line1_rep = np.random.choice(dsl._grammar['S'], p=[0.4, 0.6])
# if line1_rep == dsl._grammar['S'][0]:
# line1_rep = ": " + line1_rep
line1 = line1.replace('S', line1_rep)
# print(line1)
# line2 = line1.replace('B', np.random.choice(dsl._grammar['B'], p=[0.5, 0.5]))
# print(line2)
line2_list = line1.split(' ')
# print(line2_list)
for line2_index in range(len(line2_list)):
if line2_list[line2_index] == 'B':
line2_list[line2_index] = np.random.choice(dsl._grammar['B'],
p=[0.5, 0.5])
# print(' '.join(line2_list))
line2 = ' '.join(line2_list)
# print(line2)
line3_list = line2.split(' ')
print(line3_list)
for line3_index in range(len(line3_list)):
if line3_list[line3_index] == 'B1':
line3_list[line3_index] = np.random.choice(dsl._grammar['B1'])
# print(' '.join(line3_list))
print(line3_list)
for i in range(1, len(line3_list)):
if (line3_list[i] in ['and', 'if']):
line3_list[i] = ''
print(' '.join(line3_list))
line3 = ' '.join(line3_list)
line4_list = line3.split(' ')
# print(line4_list)
for i in range(len(line4_list)):
if (line4_list[i] == "NUMBER"):
line4_list[i] = np.random.choice(dsl._grammar['NUMBER'])
# print(' '.join(line4_list))
for i in range((len(line4_list))):
if (line4_list[i] == "SMALL_NUMBER"):
line4_list[i] = np.random.choice(dsl._grammar['SMALL_NUMBER'])
line4 = ' '.join(line4_list)
line4 = line4.replace('if', '')
# line5_list = line4.split('and')
print(line4)
rules_list = []
def get_action(self, state):
actions = state.available_moves()
for a in actions:
if DSL.isStopAction(a) and DSL.hasWonColumn(state,a):
return a
if DSL.containsNumber(a, 4) and DSL.isDoubles(a):
return a
if DSL.containsNumber(a, 2) and DSL.containsNumber(a, 6):
return a
if DSL.containsNumber(a, 2) and DSL.containsNumber(a, 4):
return a
if DSL.containsNumber(a, 6) and DSL.containsNumber(a, 4):
return a
if DSL.containsNumber(a, 2) and DSL.isDoubles(a):
return a
if DSL.containsNumber(a, 6) and DSL.isDoubles(a):
return a
if DSL.containsNumber(a, 4):
return a
if DSL.containsNumber(a, 2):
return a
if DSL.containsNumber(a, 6):
return a
return actions[0]
