Shapes and Dimensions in PyTorch

Similarly to NumPy arrays, a tensor's shape determines its dimensions. You can inspect a tensor's shape using the .shape attribute:


              123
            
import torch
tensor = torch.tensor([[1, 2, 3], [4, 5, 6]])
print(f"Tensor shape: {tensor.shape}")

Reshaping Tensors with view

The .view() method creates a new view of the tensor with the specified shape without modifying the original tensor. The total number of elements must remain the same.


              1234567
            
import torch
tensor = torch.arange(12)
# Reshape a tensor to 4x3
reshaped_tensor = tensor.view(4, 3)
print(f"Reshaped tensor: {reshaped_tensor}")
# Original tensor remains unchanged
print(f"Original tensor: {tensor}")

Reshaping Tensors with reshape

The .reshape() method is similar to .view() but can handle cases where the tensor is not stored contiguously in memory. It also doesn't modify the original tensor.


              12345
            
import torch
tensor = torch.arange(12)
# Reshape a tensor to 6x2
reshaped_tensor = tensor.reshape(6, 2)
print(f"Reshaped tensor: {reshaped_tensor}")

Using Negative Dimensions

You can use -1 in the shape to let PyTorch infer the size of one dimension based on the total number of elements.


              12345
            
import torch
tensor = torch.arange(12)
# Automatically infer the second dimension
inferred_tensor = tensor.view(2, -1)
print("Inferred Tensor:", inferred_tensor)

Understanding Tensor Views

A view of a tensor shares the same data with the original tensor. Changes to the view affect the original tensor and vice versa.


              1234567
            
import torch
tensor = torch.arange(12)
view_tensor = tensor.view(2, 6)
view_tensor[0, 0] = 999
# Changes in the view are reflected in the original tensor
print("View Tensor:", view_tensor)
print("Original Tensor:", tensor)

Changing Dimensions

The following two methods allow you to add or remove dimensions:

unsqueeze(dim) adds a new dimension at the specified position;
squeeze(dim) removes dimensions of size 1.


              12345678
            
import torch
tensor = torch.arange(12)
# Add a new dimension
unsqueezed_tensor = tensor.unsqueeze(0)  # Add a batch dimension
print(f"Unsqueezed tensor: {unsqueezed_tensor.shape}")
# Remove a dimension of size 1
squeezed_tensor = unsqueezed_tensor.squeeze(0)
print(f"Squeezed Tensor: {squeezed_tensor.shape}")

Все було зрозуміло?

Дякуємо за ваш відгук!

Секція 1. Розділ 9

Запитати АІ

Запитайте про що завгодно або спробуйте одне із запропонованих запитань, щоб почати наш чат

Зміст курсу

PyTorch Essentials