RuntimeError: the derivative for 'unique_dim' is not implemented.

1

When I added a small part of my own processing based on other people’s code, this error occurred (it seems to be an error in the process of loss_backward()), which part should I modify?
The complete error is as follows:

Traceback (most recent call last): File "D:\PyCharm 2020.1.1\plugins\python\helpers\pydev\pydevd.py", line 1438, in _exec pydev_imports.execfile(file, globals, locals) # execute the script File "D:\PyCharm 2020.1.1\plugins\python\helpers\pydev\_pydev_imps\_pydev_execfile.py", line 18, in execfile exec(compile(contents+"\n", file, 'exec'), glob, loc) File "F:/杂七杂八的代码/EntityMatcher-master/EntityMatcher-master/run.py", line 74, in <module> run_experiment(model_name, dataset_dir, embedding_dir) File "F:/杂七杂八的代码/EntityMatcher-master/EntityMatcher-master/run.py", line 32, in run_experiment model.run_train(train, File "D:\anaconda3\envs\dm\lib\site-packages\deepmatcher\models\core.py", line 183, in run_train return Runner.train(self, *args, **kwargs) File "D:\anaconda3\envs\dm\lib\site-packages\deepmatcher\runner.py", line 338, in train Runner._run( File "D:\anaconda3\envs\dm\lib\site-packages\deepmatcher\runner.py", line 249, in _run loss.backward() File "D:\anaconda3\envs\dm\lib\site-packages\torch\_tensor.py", line 255, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "D:\anaconda3\envs\dm\lib\site-packages\torch\autograd\__init__.py", line 147, in backward Variable._execution_engine.run_backward( RuntimeError: the derivative for 'unique_dim' is not implemented.

2

Based on the error message it seems you are calling torch.unique(input, dim) in your code and expect to backpropagate through it. Since this operation is not differentiable it will fail:

x = torch.randn(10, 10, requires_grad=True)
out = torch.unique(x, dim=1)
out.mean().backward()
# NotImplementedError: the derivative for 'unique_dim' is not implemented.
3

Thanks for the answer my problem was solved!