def regret(self, action):
Ew = {arm: self.theta.dot(self.xt[arm]) for arm in self.arms}
opt = self.oracle(Ew, *self.params(None))
p = [self.theta.dot(self.xt[arm]) / 2 + 0.5 for arm in action]
popt = [self.theta.dot(self.xt[arm]) / 2 + 0.5 for arm in opt]
regret = ereward(popt, self.gamma, self.disj) - ereward(p, self.gamma, self.disj)
return ereward(popt, self.gamma, self.disj) - ereward(p, self.gamma, self.disj)
def regret(self, action):
Ew = {arm: self.theta.dot(self.xt[arm]) + self.b for arm in self.arms}
opt = self.oracle(Ew, *self.params(True))
p = [(self.theta.dot(self.xt[arm]) + self.b) / 2 + 0.5 for arm in action]
#p = [self.theta.dot(self.xt[arm]) + self.b for arm in action]
popt = [(self.theta.dot(self.xt[arm]) + self.b) / 2 + 0.5 for arm in opt]
#popt = [self.theta.dot(self.xt[arm]) + self.b for arm in opt]
return ereward(popt, self.gamma, self.disj) - ereward(p, self.gamma, self.disj)
def isp_Coracle(U, G, u, v, k, gamma):
for (x, y), _ in U.items():
G[x][y]['weight'] = max(1, 0)
s = {}
for path in islice(networkx.shortest_simple_paths(G, u, v), k):
sarms = tuple(tuple(sorted((path[i], path[i+1]))) for i in range(len(path) - 1))
psarms = [(U[e] + 1) / 2 for e in sarms]
s[sarms] = ereward(psarms, gamma, False)
return list(max(s, key=s.get))
