Author | chriiscardozo |
Submission date | 2016-12-28 21:50:24.943248 |
Rating | 4501 |
Matches played | 399 |
Win rate | 43.36 |
Use rpsrunner.py to play unranked matches on your computer.
import random
from random import randint
class RAM:
def __init__(self, addressSize, entrySize, addressing, decay=None, up=None):
self.addressing = addressing
self.decay = decay
self.up = up
self.ram = {}
self.address = [ randint(0, entrySize-1) for x in range(addressSize) ]
def _addressToIndex(self, entry):
binCode = []
for i in self.address:
binCode.append(entry[i])
return self.addressing(binCode)
def _acumulateRam(self, index):
if index not in self.ram:
self.ram[index] = 0
self.ram[index] += 1
def _getValue(self, index):
if index not in self.ram:
return 0
else:
return self.ram[index]
def train(self, entry, negative=False):
index = self._addressToIndex(entry)
if not negative:
if self.up is None:
self._acumulateRam(index)
else:
self.up(entry=entry, ram=self.ram, address=self.address, index=index)
else:
self.decay(entry=entry, ram=self.ram, address=self.address, index=index)
def classify(self, entry):
index = self._addressToIndex(entry)
return self._getValue(index)
class Discriminator:
def __init__(self, name, entrySize, addressSize, addressing, numberOfRAMS=None, decay=None, up=None):
if numberOfRAMS is None:
numberOfRAMS = int(entrySize/addressSize)
self.rams = [ RAM(addressSize, entrySize, addressing, decay=decay, up=up) for x in range(numberOfRAMS) ]
def train(self, entry, negative=False):
for ram in self.rams:
ram.train(entry, negative)
def classify(self, entry):
return [ ram.classify(entry) for ram in self.rams ]
class Addressing:
def __call__(self, binCode): # binCode is a list of values of selected points of entry
index = 0
for i,e in enumerate(binCode):
if e > 0:
index += pow(2,i)
return index
class WiSARD:
def __init__(self,
addressSize = 3,
numberOfRAMS = None,
bleachingActivated = True,
seed = random.randint(0, 1000000),
sizeOfEntry = None,
classes = [],
verbose = True,
addressing = Addressing(),
makeBleaching = None,
decay = None,
up = None):
self.seed = seed
self.decay = decay
self.up = up
random.seed(seed)
if addressSize < 3:
self.addressSize = 3
else:
self.addressSize = addressSize
self.numberOfRAMS = numberOfRAMS
self.discriminators = {}
self.bleachingActivated = bleachingActivated
self.addressing = addressing
if makeBleaching is None:
self.makeBleaching = self._makeBleaching
else:
self.makeBleaching = makeBleaching
if sizeOfEntry is not None:
for aclass in classes:
self.discriminators[aclass] = Discriminator(
aclass, sizeOfEntry, self.addressSize,
self.addressing, self.numberOfRAMS, decay=self.decay, up=self.up)
def _makeBleaching(self, discriminatorsoutput):
bleaching = 0
ambiguity = True
biggestVote = 2
while ambiguity and biggestVote > 1:
bleaching += 1
biggestVote = None
ambiguity = False
for key in discriminatorsoutput:
discriminator = discriminatorsoutput[key]
limit = lambda x: 1 if x >= bleaching else 0
discriminator[1] = sum(map(limit, discriminator[0]))
if biggestVote is None or discriminator[1] > biggestVote:
biggestVote = discriminator[1]
ambiguity = False
elif discriminator[1] == biggestVote:
ambiguity = True
if self.bleachingActivated:
break
return discriminatorsoutput
def train(self, entries, classes):
sizeOfEntry = len(entries[0])
for i,entry in enumerate(entries):
aclass = str(classes[i])
if aclass not in self.discriminators:
self.discriminators[aclass] = Discriminator(
aclass, sizeOfEntry, self.addressSize,
self.addressing, self.numberOfRAMS, decay=self.decay, up= self.up)
self.discriminators[aclass].train(entry)
if self.decay is not None:
for key in self.discriminators:
if key != aclass:
self.discriminators[key].train(entry, negative=True)
def classifyEntry(self, entry):
discriminatorsoutput = {}
for keyClass in self.discriminators:
discriminatorsoutput[keyClass] = [self.discriminators[keyClass].classify(entry),0]
discriminatorsoutput = self.makeBleaching(discriminatorsoutput)
calc = lambda key: (key, float(discriminatorsoutput[key][1])/len(discriminatorsoutput[key][0]))
classes = list(map(calc,discriminatorsoutput))
classes.sort(key=lambda x: x[1], reverse=True)
return classes
def classify(self, entries):
output=[]
for i,entry in enumerate(entries):
aclass = self.classifyEntry(entry)[0][0]
output.append((entry, aclass))
return output
class Decay:
def __call__(self, **kwargs):
if kwargs['index'] in kwargs['ram']:
value = kwargs['ram'][kwargs['index']]
kwargs['ram'][kwargs['index']] = 0.5*value - 0.1
class Up:
def __call__(self, **kwargs):
if kwargs['index'] not in kwargs['ram']:
kwargs['ram'][kwargs['index']] = 1
else:
value = kwargs['ram'][kwargs['index']]
kwargs['ram'][kwargs['index']] = 0.5*value + 1
class MakeBleaching:
def __call__(self, discriminatorsoutput):
for key in discriminatorsoutput:
ramsoutput = discriminatorsoutput[key][0]
discriminatorsoutput[key][1] = sum(ramsoutput)
return discriminatorsoutput
def get_x(total, scale):
x = []
turns = sum(total)
for t in total:
filled = 0 if turns == 0 else ((float(t)/turns) * scale)
for i in range(scale):
x.append(1 if filled > i else 0)
return x
if input == "":
turn = 0
total = [0,0,0]
history = []
# hyperparameters
doubt_value = 0.3
history_size = 25
thermometer_scale = 10
min_random_turns = 0
final_random_turns = 0
address_size=8
# ****************
count = 0
wisard = WiSARD(addressSize=address_size,sizeOfEntry=3*thermometer_scale,verbose=False,seed=42,decay=Decay(),up=Up())
char2num = {"R":0, "P":1, "S":2}
num2char = {0:"R", 1:"P", 2:"S"}
defeated_by = {"R":"P", "P":"S", "S":"R"}
output = random.choice(["R","P","S"])
elif turn < min_random_turns or turn > 1000 - final_random_turns:
output = random.choice(["R","P","S"])
total[char2num[input]] += 1
history.append(char2num[input])
else:
# using thermometer method to Binarize input
x = get_x(total, thermometer_scale)
# training
wisard.train([x], [input])
total[char2num[input]] += 1
history.append(char2num[input])
new_x = get_x(total, thermometer_scale)
# classifying
result = wisard.classifyEntry(new_x)
output = defeated_by[result[0][0]]
if(turn % 10 == 0 or result[0][1] < doubt_value):
count += 1
output = random.choice(["R","P","S"])
if(len(history) > history_size):
total[history[0]] -= 0 if total[history[0]] == 0 else 1
history.pop(0)
last_prediction = output
turn += 1