y_preds.squeeze() In [18]: of "02. PyTorch Neural Network Classification"

[yabool2001]

Hi,

I've noticed that y_preds.squeeze() In [18]: of https://www.learnpytorch.io/02_pytorch_classification/ doesn't change anything. Idea was to get rid of extra dimension. But before torch.Size is ([5, 1]) and after torch.Size is ([5, 1]). Am I right or have I missed something?

[rwenen]

Hi @yabool2001
I use tensor.squeeze(), which can change the tensor's shape.
The following are the results before and after I used tensor.squeeze() to y_preds:

Hope to help you :)

[yabool2001]

Hi fooooooox,

Thx for interacting. I investigated your answer and made additional test. I see now that tensor.squeeze() can't change the tensor's shape. To be completely clear it can "returns a tensor with all specified dimensions of input of size 1 removed" (taken from pytorch docs). But it doesn't change tensor's shape. If you add

print ( f"\n{y_preds=}" ) , print ( f"{y_preds.shape=}" )

at the end of your code you will see that y_preds hasn't changed. The result will be:

y_preds = tensor([[1.], [1.], [1.], [1.], [0.]], grad_fn=<RoundBackward0>) y_preds.shape = torch.Size([5, 1])

Here you can find my verification:
https://github.com/yabool2001/squeeze_test/blob/master/squeeze_test.ipynb

Look at a_preds tensor that is proper example of getting rid of extra dimension.

Looking forward to hearing your comments!

[rwenen]

I think I know what you mean
y.squeeze() will not change to y itself unless you overwrite itself
But I think can use y.squeeze_() which will in-place the original y
(https://pytorch.org/docs/stable/generated/torch.Tensor.squeeze_.html#torch.Tensor.squeeze_)
Following is an example of what I tried:

Hope to help you :)

[yabool2001]

Great help,

squeeze_() removes the extra dimension. Like y=y.squeeze() does.

Thx again