41304b258afa59f0cc47b96dbe1af9e75b2c90e3,tests/test_dqn_head.py,_TestDQNHead,test_policy,#_TestDQNHead#,92

Before Change

            opt.update()
            // b
            pi.zerograds()
            loss = F.softmax_cross_entropy(
                pi.forward(b),
                chainer.Variable(np.asarray([1], dtype=np.int32)))
            loss.backward()
            opt.update()

        pa = float(pi(a, [0]).data)

After Change

            loss_b.backward()
            opt.update()

        pa = float(pi(head(a)).probs.data[0, 0])
        pb = float(pi(head(b)).probs.data[0, 1])
        self.assertAlmostEqual(pa, 1.0, places=3)
        self.assertAlmostEqual(pb, 1.0, places=3)

Instances