1c51cecde6aba5e6aa61158dd441326da5bb79a4,hypergan/trainers/evolution_trainer.py,EvolutionTrainer,_create,#EvolutionTrainer#,11

Before Change

        self.update_parent = update_parent
        f_lambda = config.f_lambda or 1

        grads = tf.gradients(loss.g_loss, d_vars)
        grad_d = tf.add_n([tf.reshape(gan.ops.squash(g), [1]) for g in grads])
        fq = loss.g_loss
        //TODO measure each g
        //self.measure_g = fq + f_lambda * fd
        //fd = -tf.log(grad_d - tf.log(tf.nn.sigmoid(loss.d_loss)) - tf.log(1-tf.nn.sigmoid(loss.g_loss)))

After Change

        def _squash(grads):
            return tf.add_n([tf.reshape(gan.ops.squash(g), [1]) for g in grads])
        children_grads = [_squash(tf.gradients(l, d_vars)) for l in loss.children_losses]
        if config.fitness == "g":
            self.measure_g = [-l for l in loss.children_losses]
        else:
            self.measure_g = [-l+f_lambda*(-tf.log(TINY+grad_d - tf.log(TINY+tf.nn.sigmoid(loss.d_loss)) - tf.log(TINY+1-tf.nn.sigmoid(l)))) for l, grad_d in zip(loss.children_losses, children_grads)]
        loss.metrics["measure_g"] = tf.reduce_mean(self.measure_g)
        loss.metrics["g_loss"] = loss.g_loss
        loss.metrics["d_loss"] = loss.d_loss

Instances