🐛 Bug

Tensor.std(0) returns a scalar instead of a vector.

To Reproduce

Steps to reproduce the behavior:

float data[] = { 1, 2, 3, 4, 5, 6 };
torch::Tensor f = torch::from_blob(data, {2, 3});
std::cout << f.mean(0) << std::endl;
// yields a 1-d tensor as expected.
// 2.5000
// 3.5000
// 4.5000
// [ CPUFloatType{3} ]
std::cout << f.std(0) << std::endl;
// yields a scalar, unexpected.
// 1.70783
// [ CPUFloatType{} ]

Expected behavior

Tensor.std(0) should return a vector to be consistent with other methods. As @ptrblck pointed out in this thread https://discuss.pytorch.org/t/tensor-std-0-returns-a-scalar-instead-of-a-vector/85974, the reason this is happening is because 0 is being interpreted as the unbiased argument here, and suggests f.std(true, 0) as a workaround. However, even f.std(true, 0) actually collapses dimension 1, which is also unintuitive.

Ideally, these behaviors should be fixed or listed in the documentation.

Environment

Libtorch 1.4.0 on XCode 11.4

Additional context

cc @ezyang @gchanan @zou3519 @yf225 @glaringlee @bhosmer @smessmer @ljk53 @ssnl