Function Differences with torch.var_mean

torch.var_mean

torch.var_mean(input, dim, unbiased=True, keepdim=False, *, out=None)

For more information, see torch.var_mean.

mindspore.ops.var_mean

mindspore.ops.var_mean(input, axis=None, ddof=0, keepdims=False)

For more information, see mindspore.ops.var_mean.

Differences

PyTorch: Output the variance and mean value of the Tensor in each dimension, or the variance and mean value of the specified dimension according to dim. If unbiased is True, use Bessel for correction; if False, use bias estimation to calculate the variance. keepdim controls whether the output and input dimensions are the same.

MindSpore: Output the variance and mean value of the Tensor in each dimension, or the variance and mean value of the specified dimension according to axis. If ddof is a boolean, it has the same effect as unbiased; if ddof is an integer, the divisor used in the calculation is N-ddof, where N denotes the number of elements. keepdim controls whether the output and the input have the same dimensionality.

Categories	Subcategories	PyTorch	MindSpore	Differences
Parameters	Parameter 1	input	input	Same function, different parameter names
	Parameter 2	dim	axis	Same function, different parameter names
	Parameter 3	unbiased	ddof	`ddof` is the same as `unbiased` when it is a boolean value
	Parameter 4	keepdim	keepdims	Same function, different parameter names
	Parameter 5	out	-	MindSpore does not have this parameter

Code Example# PyTorch
import torch

input = torch.tensor([[[9, 7, 4, -10],
                       [-9, -2, 1, -2]]], dtype=torch.float32)
print(torch.var_mean(input, dim=2, unbiased=True, keepdim=True))
# (tensor([[[73.6667],
#          [18.0000]]]), tensor([[[ 2.5000],
#          [-3.0000]]]))

# MindSpore
import mindspore as ms

input = ms.Tensor([[[9, 7, 4, -10],
                    [-9, -2, 1, -2]]], ms.float32)
print(ms.ops.var_mean(input, axis=2, ddof=True, keepdims=True))
# (Tensor(shape=[1, 2, 1], dtype=Float32, value=
# [[[ 7.36666641e+01],
#   [ 1.79999981e+01]]]), Tensor(shape=[1, 2, 1], dtype=Float32, value=
# [[[ 2.50000000e+00],
#   [-3.00000000e+00]]]))