无法弄清楚 pytorch 代码中的就地操作？

Question

I have the following implementation in PyTorch for learning using LSTM:

https://gist.github.com/rahulbhadani/f1d64042cc5a80280755cac262aa48aa

However, the code is experiencing in-place operation error

where my error output is:

/home/ivory/anaconda3/lib/python3.7/site-packages/ipykernel_launcher.py:10: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  # Remove the CWD from sys.path while we load stuff.
---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-86-560ec78f2b64> in <module>
     27 linear = torch.nn.Linear(hidden_nums, output_dim)
     28 
---> 29 global_loss_list = global_training(lstm2)

<ipython-input-84-152890a3028c> in global_training(optimizee)
      3     adam_global_optimizer = torch.optim.Adam([{'params': optimizee.parameters()}, 
      4                                      {'params':linear.parameters()}], lr = 0.0001)
----> 5     _, global_loss_1 = learn2(LSTM_Optimizee, training_steps, retain_graph_flag=True, reset_theta=True)
      6 
      7     print(global_loss_1)

<ipython-input-83-0357a528b94d> in learn2(optimizee, unroll_train_steps, retain_graph_flag, reset_theta)
     43             # requires_grad=True. These are accumulated into x.grad for every
     44             # parameter x. In pseudo-code: x.grad += dloss/dx
---> 45             loss.backward(retain_graph = retain_graph_flag) #The default is False, when the optimized LSTM is set to True
     46 
     47             print('x.grad: {}'.format(x.grad))

~/anaconda3/lib/python3.7/site-packages/torch/tensor.py in backward(self, gradient, retain_graph, create_graph)
    116                 products. Defaults to ``False``.
    117         """
--> 118         torch.autograd.backward(self, gradient, retain_graph, create_graph)
    119 
    120     def register_hook(self, hook):

~/anaconda3/lib/python3.7/site-packages/torch/autograd/__init__.py in backward(tensors, grad_tensors, retain_graph, create_graph, grad_variables)
     91     Variable._execution_engine.run_backward(
     92         tensors, grad_tensors, retain_graph, create_graph,
---> 93         allow_unreachable=True)  # allow_unreachable flag
     94 
     95 

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.FloatTensor [1, 10]] is at version 1; expected version 0 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

I have tried to trace the error but didn't succeed. Any help in this regard will be appreciated.

Thanks.

Answer 1

I think the issue is with the following line:

global_loss_list.append(global_loss.detach_())

The convention in PyTorch for in-place operations is using _ at the end of the function name (as in detach_ ). I believe you shouldn't be detaching in-place. In other words, change detach_ to detach