.multinomial()

In PyTorch, the .multinomial() function is used to sample elements from a given probability distribution. It is commonly used when selecting multiple indices based on given probabilities, such as in reinforcement learning or probabilistic modeling.

Syntax

torch.multinomial(input, num_samples, replacement=False, *, gen=None, out=None)

• input: A tensor containing probabilities for each element. The values must be non-negative.
• num_samples: The number of samples to draw from the probability distribution.
• replacement (Optional): If False, the same element is not chosen more than once.
• gen (Optional): A random number generator for reproducibility.
• out (Optional): The output tensor to store results.

Example

The following example demonstrates the usage of the .multinomial() function:

import torch

# Define a tensor containing probabilities
ten = torch.tensor([0.1, 0.3, 0.4, 0.2])

# Sample 2 indices without replacement
res = torch.multinomial(ten, num_samples=2, replacement=False)

# Print the result
print(res)

The above code produces the following output:

tensor([3, 2])

Note: Since elements are chosen randomly, the output may vary each time the code is run.