你好关于tsne可视化，提供的代码 这块的代码不对吧？

[yhl2018]

for batch_idx, (input1, input2, label1, label2) in enumerate(trainloader):
labels = torch.cat((label1, label2), 0)
z1 = torch.ones(label1.shape)
z2 = torch.zeros(label2.shape)
z = torch.cat((z1, z2), 0)
print(batch_idx)
input1 = Variable(input1.cuda())
input2 = Variable(input2.cuda())

      a = labels.unique()
      for i in range(len(a)):
        for j in range(len(labels)):
          if labels[j] == a[i]:
              labels[j] = i
      #print(labels)
      data_time.update(time.time() - end)
      out = torch.cat((input1, input2), 0)

      tsne = manifold.TSNE(n_components=2, init='pca', random_state=0)
      X_tsne = tsne.fit_transform(out.detach().cpu().numpy())
      plot_embedding(X_tsne, labels, z)
      plt.savefig(osp.join('save_tsne', 'tsne_{}.jpg'.format(batch_idx)))