def estimate_internal(self, input, output): pr = Primitives() out = tf.reduce_sum(input) out += self.rho ret = pr.and_diff(pr.geq(out, output[0]), pr.geq(output[1], out)) return ret
def estimate_internal(self, input, output): pr = Primitives() best = -pr.get_inf() best_by_index = tf_const(0) for i in range(0, self.array_size): d = input[i] + self.noise[:, i] hit = pr.eq(tf_const(i), output) best = pr.ite(hit, best, tf.maximum(best, d)) best_by_index = pr.ite(hit, input[i], best_by_index) rate = self.params.epsilon / 2 x = best - best_by_index return pr.ite(pr.geq(x, tf_const(0)), exponential_cdf_right(rate, x), tf_const(1))
def estimate_internal(self, input, output):
pr = Primitives()
best = -pr.get_inf()
best_by_index = tf_const(0)
for i in range(0, self.array_size):
d = input[i] + self.noise[:, i]
hit = pr.eq(tf_const(i), output)
best = pr.ite(hit, best, tf.maximum(best, d))
best_by_index = pr.ite(hit, input[i], best_by_index)
scale = 2 / self.params.epsilon
x = best - best_by_index
return tf.exp(laplacian_cdf(x, scale)[1])
def estimate_internal(self, input, output):
pr = Primitives()
out = []
for i in range(self.params.array_size):
above_threshold = pr.geq(input[i],
self.params.threshold + self.rho)
hit = above_threshold
out.append(pr.ite(hit, 1, 0))
ret = 1
for i in range(self.params.array_size):
ret = pr.and_diff(ret, pr.eq(out[i], output[i]))
return ret
def estimate_internal(self, input, output):
pr = Primitives()
count = tf_const(0)
c = tf_const(self.params.c)
out = []
for i in range(self.params.array_size):
above_threshold = pr.geq(input[i] + self.nu[:, i],
self.params.threshold + self.rho)
abort = pr.eq(count, c)
hit = pr.and_diff(above_threshold, pr.not_diff(abort))
count = pr.ite(hit, count + 1, count)
out.append(pr.ite(hit, 1, 0))
ret = 1
for i in range(self.params.array_size):
ret = pr.and_diff(ret, pr.eq(out[i], output[i]))
return ret