Скачать или смотреть How to Create a Mutable Tensor in PyTorch

Learn how to create a `mutable tensor` in PyTorch that reflects changes to the original tensor, including code examples and detailed explanations.
---
This video is based on the question https://stackoverflow.com/q/67183122/ asked by the user 'Penguin' ( https://stackoverflow.com/u/14735451/ ) and on the answer https://stackoverflow.com/a/67183249/ provided by the user 'Edwin Cheong' ( https://stackoverflow.com/u/7617767/ ) at 'Stack Overflow' website. Thanks to these great users and Stackexchange community for their contributions.

Visit these links for original content and any more details, such as alternate solutions, latest updates/developments on topic, comments, revision history etc. For example, the original title of the Question was: How to create a PyTorch mutable tensor?

Also, Content (except music) licensed under CC BY-SA https://meta.stackexchange.com/help/l...
The original Question post is licensed under the 'CC BY-SA 4.0' ( https://creativecommons.org/licenses/... ) license, and the original Answer post is licensed under the 'CC BY-SA 4.0' ( https://creativecommons.org/licenses/... ) license.

If anything seems off to you, please feel free to write me at vlogize [AT] gmail [DOT] com.
---
How to Create a Mutable Tensor in PyTorch

In the world of deep learning and data manipulation, understanding how to effectively manage tensors and their changes is crucial. One common need arises when you want to create a tensor that can reflect changes made to another tensor in PyTorch. In this guide, we'll dive into this problem and explore how to create a mutable tensor that updates with changes made to the original tensor.

Understanding the Problem

The challenge you might face is straightforward: You want to create a copy of a tensor such that when the original tensor is modified, the changes are also reflected in the copied tensor. This might seem simple at first, but as many have discovered, using functions like clone() or detach() doesn't yield the expected results. Let’s examine a common attempt to solve this problem:

[[See Video to Reveal this Text or Code Snippet]]

Expected Output

You might expect both r and p to show the updated value, but when you run this code, you'll see:

[[See Video to Reveal this Text or Code Snippet]]

In this case, the original tensor r has changed, but the clone p remained the same. This occurs because clone() creates a new copy of the tensor, and so p does not keep a reference to the original data.

The Solution: Using view() for Mutable Tensors

The solution to this problem lies in utilizing the view() method in PyTorch, which creates a view of the original tensor. A view is a shallow copy, meaning that it references the same data. Here’s how you can create a mutable tensor:

Step-by-step Example

Create the Original Tensor:
First, create the original tensor r as follows:

[[See Video to Reveal this Text or Code Snippet]]

Create a Mutable Tensor:
Next, create a view of the tensor r using view():

[[See Video to Reveal this Text or Code Snippet]]

Modify the Original Tensor:
When you modify r, it will automatically update p as both variables will reference the same data:

[[See Video to Reveal this Text or Code Snippet]]

Important Notes

The use of with torch.no_grad(): is important to ensure that the gradient tracking does not interfere with the mathematical operations you perform on the tensor.

Using in-place operations (like mul_()) ensures that you modify the tensor data directly without reassigning it to a new variable.

Conclusion

With this method, you can effectively create a mutable tensor in PyTorch that accurately reflects changes made to its original counterpart. By using the view() method coupled with careful management of gradients, you maintain a link between your tensors that is critical for many applications in deep learning and tensor manipulation.

Feel free to experiment with this approach in your PyTorch projects and enjoy the seamless tensor manipulation it offers!