ef234844e1604d5496607fcf48886d9e556c0685,engine/training.py,,run,#Any#Any#Any#Any#,20

Before Change

        apply_grad_op = train_step.apply_gradients(ave_grads)
        // summary for visualisations
        // tracking current batch loss
        summaries = [tf.summary.scalar("total-miss", ave_miss),
                     tf.summary.scalar("total-loss", ave_loss)]

        // Track the moving averages of all trainable variables.

After Change

                //     current_iter, loss_value, miss_value, iter_time))
                print("iter {:d}, loss={:.8f} ({:.3f}s)".format(
                    current_iter, loss_value, iter_time))
                if (current_iter % 20) == 0:
                    writer.add_summary(sess.run(write_summary_op), current_iter)
                    // Plot reconstructions for the basic autoencoder
                    for p in range(0,4):
                        plt.subplot(4, 2, 2*p+1)
                        temp1 = sess.run(predictions[0])
                        temp1 = temp1[p,:,12,:,0]
                        temp1.reshape(24, 24)
                        plt.imshow(temp1, cmap="gray")
                        plt.subplot(4, 2, 2*p+2)
                        temp2 = sess.run(predictions[1])
                        temp2 = temp2[p, :, 12, :, 0]
                        temp2.reshape(24, 24)
                        plt.imshow(temp2, cmap="gray")
                    plt.pause(0.0001)
                if (current_iter % param.save_every_n) == 0 and i > 0:
                    saver.save(sess, ckpt_name, global_step=current_iter)
                    print("Iter {} model saved at {}".format(
                        current_iter, ckpt_name))

Instances