SelfO0-6_mix_v3 (fixed)

Use rpsrunner.py to play unranked matches on your computer.

Source code:

#By David Catt 2013.
#Takes its own output as context.
import random
RPSID = ["R", "P", "S"]
RPSWIN = [1, 2, 0]
RPSLOSE = [2, 0, 1]
IMID = 2147483648
IMAX = 4294967295
class RPSMDL:
    def __init__(self, order, decay, init):
        self.mod = 3**order
        self.ctx = 0
        self.rate = decay
        self.mdl = [[float(init), float(init), float(init)] for a in range(self.mod)]
    def predict(self):
        return self.mdl[self.ctx]
    def update(self, val):
        val %= 3
        #self.mdl[self.ctx][val] += (IMAX - self.mdl[self.ctx][val]) >> self.rate
        #self.mdl[self.ctx][RPSWIN[val]] -= self.mdl[self.ctx][RPSWIN[val]] >> (self.rate + 1)
        #self.mdl[self.ctx][RPSLOSE[val]] -= self.mdl[self.ctx][RPSLOSE[val]] >> (self.rate + 1)
        self.mdl[self.ctx][0] *= self.rate
        self.mdl[self.ctx][1] *= self.rate
        self.mdl[self.ctx][2] *= self.rate
        self.mdl[self.ctx][val] += 1
        self.ctx = ((self.ctx * 3) + val) % self.mod;
        
class RPSSMDL:
    def __init__(self, order, decay, init):
        self.mod = 3**order
        self.ctx = 0
        self.rate = decay
        self.mdl = [[float(init), float(init), float(init)] for a in range(self.mod)]
    def predict(self):
        return self.mdl[self.ctx]
    def update(self, last, val):
        last %= 3
        val %= 3
        #self.mdl[self.ctx][val] += (IMAX - self.mdl[self.ctx][val]) >> self.rate
        #self.mdl[self.ctx][RPSWIN[val]] -= self.mdl[self.ctx][RPSWIN[val]] >> (self.rate + 1)
        #self.mdl[self.ctx][RPSLOSE[val]] -= self.mdl[self.ctx][RPSLOSE[val]] >> (self.rate + 1)
        self.mdl[self.ctx][0] *= self.rate
        self.mdl[self.ctx][1] *= self.rate
        self.mdl[self.ctx][2] *= self.rate
        self.mdl[self.ctx][val] += 1
        self.ctx = ((self.ctx * 3) + last) % self.mod;


class RPSPREDICTOR:
    def __init__(self, maxord, adapt):
        self.models = [[RPSMDL(0, 0, 0), RPSSMDL(0, 0, 0)] for a in range(maxord + 1)]
        self.weights = [[IMID, IMID] for a in range(maxord + 1)]
        self.last = [[0, 0] for a in range(maxord + 1)]
        self.maxord = maxord
        self.rate = adapt
        self.prev = 0
        for o in range(0, self.maxord):
            self.models[o][0] = RPSMDL(o, 1 - ((1.3**o) * 0.04), o + 1)
            self.models[o][1] = RPSSMDL(o, 1 - ((1.3**o) * 0.04), o + 1)
    def predict(self):
        tmp = [0.0, 0.0, 0.0]
        tot = 0
        ratios = [0.0, 0.0, 0.0]
        for o in range(0, self.maxord):
            for s in range(0, 1):
                tmp = self.models[o][s].predict()
                tot = tmp[0] + tmp[1] + tmp[2]
                if tot == 0.0:
                    tot = 1.0
                ratios[0] += (tmp[0] * self.weights[o][s] * (1.5**o)) / tot
                ratios[1] += (tmp[1] * self.weights[o][s] * (1.5**o)) / tot
                ratios[2] += (tmp[2] * self.weights[o][s] * (1.5**o)) / tot
                if tmp[0] > tmp[1]:
                    if tmp[0] > tmp[2]:
                        self.last[o][s] = 0
                    else:   
                        self.last[o][s] = 2
                else:
                    if tmp[2] > tmp[1]:
                        self.last[o][s] = 2
                    elif tmp[0] == tmp[1] and tmp[1] == tmp[2]:
                        self.last[o][s] = random.choice([0, 1, 2])
                    else:
                        self.last[o][s] = 1
        if ratios[0] > ratios[1]:
            if ratios[0] > ratios[2]:
                self.prev = 1
                return 0
            else:
                self.prev = 0
                return 2
        else:
            if ratios[2] > ratios[1]:
                self.prev = 0
                return 2
            else:
                self.prev = 2
                return 1
    def update(self, val):
        for o in range(0, self.maxord):
            self.models[o][0].update(val)
            if val == self.last[o][0]:
                self.weights[o][0] += (IMAX - self.weights[o][0]) >> self.rate
            else:
                self.weights[o][0] -= self.weights[o][0] >> (self.rate + 1)
            self.models[o][1].update(self.prev, val)
            if val == self.last[o][1]:
                self.weights[o][1] += (IMAX - self.weights[o][1]) >> self.rate
            else:
                self.weights[o][1] -= self.weights[o][1] >> (self.rate + 1)

        
if input == "":
    predictor = RPSPREDICTOR(6, 4)
    aic = 0
elif input == "R":
    predictor.update(0)
elif input == "P":
    predictor.update(1)
elif input == "S":
    predictor.update(2)
aic = RPSWIN[predictor.predict()]
output = RPSID[aic]