RuntimeError: Trying to backward through the graph a second time (or directly access saved tensor...

原因：

在跑深度学习中出现：RuntimeError: Trying to backward through the graph a second time (or directly access saved tensors after they have already been freed).  Saved intermediate values of the graph are freed when you call .backward() or autograd.grad().  Specify retain_graph=True if you need to backward through the graph a second time or if you need to access saved tensors after calling backward.

解决办法：

发现 retain_graph=True 也不对，原因是我想要将每次跑出的结果加入下一次训练迭代，直接将预测的结果y_pre赋给x，改变了tensor的梯度（grad），其实只需要改变data属性就行，错误结果为：

xx = y_pred

修改正确为：

xx = y_pred.data