Differences with torch.multinomial

The following mapping relationships can be found in this file.

PyTorch APIs	MindSpore APIs
torch.multinomial	mindspore.ops.multinomial
torch.Tensor.multinomial	mindspore.Tensor.multinomial

torch.multinomial

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

For more information, see torch.multinomial.

mindspore.ops.multinomial

mindspore.ops.multinomial(input, num_samples, replacement=True, seed=None)

For more information, see mindspore.ops.multinomial.

Differences

API function of MindSpore is consistent with that of PyTorch.

MindSpore: The default value of the parameter replacement is True , which means the sampled data is put back after each sampling.

PyTorch: The default value of the parameter replacement is False , which means the sampled data is not put back after each sampling.

Categories	Subcategories	PyTorch	MindSpore	Differences
Parameters	Parameter 1	input	input	Consistent
	Parameter 2	num_samples	num_samples	Consistent
	Parameter 3	replacement	replacement	The default value for PyTorch is False and the default value for MindSpore is True
	Parameter 4	generator	seed	For details, see General Difference Parameter Table
	Parameter 5	out	-	For details, see General Difference Parameter Table

Code Example

# PyTorch
import torch

input = torch.tensor([0, 9, 4, 0], dtype=torch.float32)
output = torch.multinomial(input, 2)
print(output)
# tensor([1, 2]) or tensor([2, 1])

# MindSpore
import mindspore as ms

input = ms.Tensor([0, 9, 4, 0], dtype=ms.float32)
output = ms.ops.multinomial(input, 2, False)
print(output)
# [1 2] or [2 1]