Creating Random Tensors

Random tensors are useful for initializing data or weights in machine learning models (the most common use case).

Random Uniform Tensors

The torch.rand() function is used to create a tensor with random values drawn from a uniform distribution between 0 and 1. Similarly to the zeros() and ones()functions, the arguments specify the shape of the tensor.


              1234
            
import torch
# Create a 6x8 tensor with random values between 0 and 1
random_tensor = torch.rand(6, 8)
print(random_tensor)

Random Normal Tensors

The torch.randn() function is used to create a tensor with random values drawn from a standard normal distribution (mean = 0, standard deviation = 1).


              1234
            
import torch
# Create a 2x2 tensor with random values from a normal distribution
normal_tensor = torch.randn(2, 2)
print(normal_tensor)

Random Integer Tensors

The torch.randint() function is used to create a tensor with random integer values drawn from discrete uniform distribution.

The first two parameters of this functions (low, which is equal to 0 by default, and high) specify the range of values (from low to high exclusive). The next parameter specifies the shape of the tensor as a tuple.


              1234
            
import torch
# Create a 4x3 tensor with random integers between 0 and 10
integer_tensor = torch.randint(0, 10, (4, 3))
print(integer_tensor)

Setting Random Seed

To ensure reproducibility, you can set a manual seed. This fixes the random numbers generated so they are the same each time you run the code.


              123456
            
import torch
# Set the random seed
torch.manual_seed(42)
# Create a 2x3 tensor with random values
seeded_tensor = torch.rand(2, 3)
print(seeded_tensor)

Practical Use Cases for Random Tensors

Weight initialization: random tensors are often used to initialize weights in neural networks;
Simulating data: generate random datasets for testing and experimentation;
Random sampling: use random tensors for tasks like dropout and noise addition in models.

Все було зрозуміло?

Дякуємо за ваш відгук!

Секція 1. Розділ 5