Reshaping (view, reshape, squeeze, unsqueeze) in PyTorch - Model Metrics & Evaluation

Metrics & Evaluation - Reshaping (view, reshape, squeeze, unsqueeze)

Which metric matters for Reshaping and WHY

When working with reshaping operations like view, reshape, squeeze, and unsqueeze in PyTorch, the key "metric" is tensor shape correctness. This means the output tensor must have the exact shape expected for the next step in your model or data pipeline.

Why? Because reshaping changes how data is organized without changing the data itself. If the shape is wrong, your model will crash or give wrong results. So, the "metric" here is not accuracy or loss but shape compatibility and data integrity.

Confusion Matrix or Equivalent Visualization

For reshaping, we don't have a confusion matrix. Instead, we can visualize shapes before and after reshaping.

    Original tensor shape: (2, 3, 4)
    After view(6, 4): (6, 4)
    After squeeze() if any dimension is 1, it is removed
    After unsqueeze(1): adds a dimension at position 1
    

Example:

    tensor = torch.randn(2, 1, 4)
    print(tensor.shape)  # (2, 1, 4)
    squeezed = tensor.squeeze()
    print(squeezed.shape)  # (2, 4)
    unsqueezed = squeezed.unsqueeze(1)
    print(unsqueezed.shape)  # (2, 1, 4)
    

Tradeoff: Flexibility vs Errors in Reshaping

Using reshape is flexible because it can handle some cases automatically, but it may copy data, which is slower.

view is faster because it returns a view without copying, but it requires the tensor to be contiguous in memory. If not, it will error.

Tradeoff: Use view for speed when you know the tensor is contiguous. Use reshape when you want safety and flexibility.

squeeze and unsqueeze help add or remove dimensions of size 1, which is useful for matching shapes without changing data.

What Good vs Bad Looks Like

Good: After reshaping, the tensor shape matches the expected input for the next layer or operation. No errors occur. Data order is preserved.

Bad: Shapes don't match, causing runtime errors. Using view on a non-contiguous tensor causes errors. Data is accidentally reordered or lost.

Example of bad:

    tensor = torch.randn(2, 3, 4).transpose(1, 2)  # non-contiguous
    tensor.view(6, 4)  # ERROR: tensor is not contiguous
    

Common Pitfalls

• Using view on a non-contiguous tensor causes errors.
• Confusing squeeze and unsqueeze positions can lead to wrong shapes.
• Assuming reshape never copies data; sometimes it does, which can affect performance.
• Not checking tensor shape before reshaping leads to silent bugs or crashes later.
• Mixing batch size and feature dimensions incorrectly during reshape.

Self Check

Your model expects input shape (batch_size, 10, 10). You have a tensor of shape (20, 100). You try tensor.view(20, 10, 10). Is this correct?

Answer: Yes, because 20*100 = 20*10*10, so the total number of elements matches. The reshape is valid if the tensor is contiguous.

But if the tensor is not contiguous, view will error. Use reshape to be safe.

Key Result

The key metric for reshaping is ensuring the output tensor shape matches expected dimensions exactly to avoid runtime errors.

Practice

1. What does the torch.squeeze() function do to a tensor?

easy

A. Adds a new dimension of size 1 at a specified position

B. Removes all dimensions of size 1 from the tensor

C. Changes the tensor's data without changing its shape

D. Flattens the tensor into a 1D vector

Solution

1. Step 1: Understand the purpose of squeeze

   torch.squeeze() removes dimensions that have size 1, effectively reducing unnecessary dimensions.

2. Step 2: Differentiate from other reshape functions

   Unlike unsqueeze which adds size 1 dims, squeeze removes them without changing data.

3. Final Answer:

   Removes all dimensions of size 1 from the tensor -> Option B

4. Quick Check:

   squeeze removes size 1 dims = B [OK]

Hint: Squeeze removes dims with size 1, like cleaning empty boxes [OK]

Common Mistakes:

• Confusing squeeze with unsqueeze
• Thinking squeeze changes data values
• Assuming squeeze flattens the tensor

2. Which of the following is the correct syntax to add a new dimension at position 0 using PyTorch?

easy

A. tensor.unsqueeze(0)

B. tensor.unsqueeze(1)

C. tensor.squeeze(0)

D. tensor.reshape(0)

Solution

1. Step 1: Recall unsqueeze syntax

   unsqueeze(dim) adds a dimension of size 1 at the specified dim index.

2. Step 2: Identify correct dimension index

   To add at position 0 (the front), use unsqueeze(0). tensor.unsqueeze(1) adds at position 1, C squeezes, D is invalid.

3. Final Answer:

   tensor.unsqueeze(0) -> Option A

4. Quick Check:

   unsqueeze(0) adds dim at front = A [OK]

Hint: unsqueeze(dim) adds size 1 dim at dim index [OK]

Common Mistakes:

• Using squeeze instead of unsqueeze
• Passing wrong dimension index
• Using reshape with a single integer argument

3. Given the tensor t = torch.randn(2, 1, 3), what is the shape after t.view(2, 3)?

medium

A. (2, 3)

B. (2, 1, 3)

C. (1, 6)

D. RuntimeError due to incompatible shape

Solution

1. Step 1: Understand original tensor shape

   Tensor t has shape (2, 1, 3), total elements = 2*1*3 = 6.

2. Step 2: Check if view(2, 3) is compatible

   New shape (2, 3) also has 6 elements, so view can reshape without error.

3. Final Answer:

   (2, 3) -> Option A

4. Quick Check:

   view reshapes without changing total elements = A [OK]

Hint: view reshapes if total elements match exactly [OK]

Common Mistakes:

• Confusing view with squeeze or unsqueeze
• Assuming view changes data order
• Trying incompatible shapes causing runtime error

4. What is wrong with this code snippet?

import torch
t = torch.randn(3, 1, 4)
t = t.view(3, 3, 1, 1)

medium

A. The new shape has more elements than original tensor

B. view cannot add new dimensions, only reshape existing ones

C. view requires contiguous tensor, but t is not contiguous

D. The new shape is incompatible with the total number of elements

Solution

1. Step 1: Calculate original tensor elements

   Original shape (3,1,4) has 3*1*4 = 12 elements.

2. Step 2: Calculate new shape elements

   New shape (3,3,1,1) has 3*3*1*1 = 9 elements, which does not match.

3. Final Answer:

   The new shape is incompatible with the total number of elements -> Option D

4. Quick Check:

   view fails if total elements don't match = C [OK]

Hint: Check total elements before view; must match exactly [OK]

Common Mistakes:

• Assuming view cannot add dimensions
• Confusing contiguous memory errors
• Miscounting total elements

5. You have a tensor x of shape (4, 1, 28, 28). You want to prepare it for a fully connected layer that expects input shape (4, 784). Which sequence of PyTorch operations correctly reshapes x without changing data order?

hard

A. x = x.view(4, 28, 28)

B. x = x.unsqueeze(1).view(4, 1, 784)

C. x = x.squeeze(1).view(4, 784)

D. x = x.reshape(4, 1, 784)

Solution

1. Step 1: Understand original tensor shape

   x has shape (4, 1, 28, 28). The middle dimension is size 1, which is unnecessary for flattening.

2. Step 2: Remove the size 1 dimension

   Use squeeze(1) to remove the dimension of size 1 at index 1, resulting in shape (4, 28, 28).

3. Step 3: Flatten the last two dimensions

   Use view(4, 784) to reshape (4, 28, 28) into (4, 784) without changing data order.

4. Final Answer:

   x = x.squeeze(1).view(4, 784) -> Option C

5. Quick Check:

   Remove size 1 dim then flatten = D [OK]

Hint: Squeeze first to remove size 1 dims, then view to flatten [OK]

Common Mistakes:

• Trying to view directly without squeezing
• Using unsqueeze unnecessarily
• Confusing reshape and view effects