【PyTorch】The difference between Tensor and Tensor

The source code of the framework listed in this article is based on PyTorch1.0, and the interactive statements are tested and passed on 0.4.1.

import torch

In PyTorch, both Tensor and Tensor can be used to generate new tensors:

>>> a=torch.Tensor([1,2])
>>> a
tensor([1., 2.])
>>> a=torch.tensor([1,2])
>>> a
tensor([1, 2])

But what is the difference between the usage of these two? I have not found suitable Chinese information, such as/t/what-is-the-difference-between-tensor-and-tensor-is-tensor-going-to-be-deprecated-in-the-future/17134/8It is also outdated, so just do it yourself and have enough food and clothing.

First of all, we need to make it clear that () is a python class, and more explicitly, it is an alias for the default tensor type (). ([1,2]) will call the constructor of Tensor class __init__ to generate a single-precision floating-point type tensor.

>>> a=torch.Tensor([1,2])
>>> a.type()
''

And () is just a python function:/docs/stable/#, the function prototype is:

torch.tensor(data, dtype=None, device=None, requires_grad=False)

Where data can be: list, tuple, NumPy ndarray, scalar and other types.
It will copy the data part in data (rather than referencing it directly), and generate the corresponding and and according to the original data type.

>>> a=torch.tensor([1,2])
>>> a.type()
''

>>> a=torch.tensor([1.,2.])
>>> a.type()
''

>>> a=np.zeros(2,dtype=np.float64)
>>> a=torch.tensor(a)
>>> a.type()
''

Let me talk about it again here (), according to/docs/stable/?highlight=empty#, we can generate tensors with specified types, specified devices and other parameters. Since () can only specify the data type, () can be regarded as a special case of ().

Finally put a small easter egg

>>> a=torch.tensor(1)
>>> a
tensor(1)
>>> a.type()
''
>>> a=torch.Tensor(1)
>>> a
tensor([0.])
>>> a.type()
''