Why define the Variable before start training ?

[SamuelQZQ]

I have noticed that, in your 'train_*.py' scipt, you always define the Variable before start training. And in every training step, you use Tensor.resize_() method and Tensor.copy_() method to prepare the Variable.

I am curious about it. What is the advantage of this trick ? Is this a trick to make the training faster ? Thank you !