Name: Set dtype with dtype argument functions and get it in PyTorch
Rating: 1.4 (960 reviews)
Author: hyperkai

Buy Me a Coffee☕

*Memos:

My post explains how to set device with device argument functions and get it.
My post explains how to set requires_grad with requires_grad argument functions and get it.
My post explains how to set keepdim with keepdim argument functions.
My post explains how to set out with out argument functions.

You can set dtype with the functions which have dtype arguments and get it with dtype and type() as shown below:

*Memos:

I selected some popular dtype argument functions such as tensor(), arange(), rand(), rand_like(), sum() and view(). *Memos:
Basically, dtype(Optional-Default:None-Type:dtype).
Basically, if dtype is None, get_default_dtype() is used or it's inferred from data, then additionally for floating point numbers, get_default_dtype() is used. *My post explains get_default_dtype() and set_default_dtype().
dtype can also accept int(), float() and bool() but not complex() which are python built-in functions.
Basically, dtype= must be used.
dtype cannot be int or float for a complex tensor.

tensor(). *My post explains tensor():

import torch

my_tensor = torch.tensor(data=[0, 1, 2], dtype=torch.float64)
my_tensor = torch.tensor(data=[0, 1, 2], dtype=float)

my_tensor, my_tensor.dtype, my_tensor.type()
# (tensor([0., 1., 2.],
#  dtype=torch.float64), torch.float64,
#  'torch.DoubleTensor')

my_tensor = torch.tensor(data=[0, 1, 2], dtype=torch.complex64)

my_tensor, my_tensor.dtype, my_tensor.type()
# (tensor([0.+0.j, 1.+0.j, 2.+0.j]),
#  torch.complex64,
#  'torch.ComplexFloatTensor')

my_tensor = torch.tensor(data=[0, 1, 2], dtype=torch.bool)
my_tensor = torch.tensor(data=[0, 1, 2], dtype=bool)

my_tensor, my_tensor.dtype, my_tensor.type()
# (tensor([False, True, True]), torch.bool, 'torch.BoolTensor')

my_tensor = torch.tensor(data=[0., 1., 2.], dtype=torch.int64)
my_tensor = torch.tensor(data=[0., 1., 2.], dtype=int)

my_tensor, my_tensor.dtype, my_tensor.type()
# (tensor([0, 1, 2]), torch.int64, 'torch.LongTensor')

my_tensor = torch.tensor(data=[0., 1., 2.], dtype=torch.complex64)

my_tensor, my_tensor.dtype, my_tensor.type()
# (tensor([0.+0.j, 1.+0.j, 2.+0.j]),
#  torch.complex64,
#  'torch.ComplexFloatTensor')

my_tensor = torch.tensor(data=[0., 1., 2.], dtype=torch.bool)
my_tensor = torch.tensor(data=[0., 1., 2.], dtype=bool)

my_tensor, my_tensor.dtype, my_tensor.type()
# (tensor([False, True, True]), torch.bool, 'torch.BoolTensor')

my_tensor = torch.tensor(data=[0.+0.j, 0.+7.j, 2.+0.j], dtype=torch.bool)
my_tensor = torch.tensor(data=[0.+0.j, 0.+7.j, 2.+0.j], dtype=bool)

my_tensor, my_tensor.dtype, my_tensor.type()
# (tensor([False, True, True]), torch.bool, 'torch.BoolTensor')

my_tensor = torch.tensor(data=[True, False, True], dtype=torch.int64)
my_tensor = torch.tensor(data=[True, False, True], dtype=int)

my_tensor, my_tensor.dtype, my_tensor.type()
# (tensor([1, 0, 1]), torch.int64, 'torch.LongTensor')

my_tensor = torch.tensor(data=[True, False, True], dtype=torch.float64)
my_tensor = torch.tensor(data=[True, False, True], dtype=float)

my_tensor, my_tensor.dtype, my_tensor.type()
# (tensor([1., 0., 1.], dtype=torch.float64),
#  torch.float64,
#  'torch.DoubleTensor')

my_tensor = torch.tensor(data=[True, False, True], dtype=torch.complex64)

my_tensor, my_tensor.dtype, my_tensor.type()
# (tensor([1.+0.j, 0.+0.j, 1.+0.j]),
#  torch.complex64,
#  'torch.ComplexFloatTensor')

arange(). *My post explains arange():

import torch

my_tensor = torch.arange(start=5, end=15, step=3, dtype=torch.float64)

my_tensor, my_tensor.dtype, my_tensor.type()
# (tensor([5., 8., 11., 14.], dtype=torch.float64),
#  torch.float64,
#  'torch.DoubleTensor')

rand(). *My post explains rand():

import torch

my_tensor = torch.rand(size=(3,), dtype=torch.float64)

my_tensor, my_tensor.dtype, my_tensor.type()
# (tensor([0.4620, 0.6369, 0.5189], dtype=torch.float64),
#  torch.float64,
#  'torch.DoubleTensor')

rand_like(). *My post explains rand_like():

import torch

my_tensor = torch.rand_like(input=torch.tensor([7., 4., 5.]), 
                            dtype=torch.float64)
my_tensor, my_tensor.dtype, my_tensor.type()
# (tensor([0.7677, 0.2914, 0.3266], dtype=torch.float64),
#  torch.float64,
#  'torch.DoubleTensor')

sum(). *My post explains sum():

import torch

my_tensor = torch.sum(input=torch.tensor([0., 1., 2., 3.]),
                      dtype=torch.float64)
my_tensor, my_tensor.dtype, my_tensor.type()
# (tensor(6., dtype=torch.float64), torch.float64, 'torch.DoubleTensor')

view(). *My post explains view():

import torch

my_tensor1 = torch.tensor([0., 1., 2.]).view(size=(3, 1))

my_tensor2 = my_tensor.view(dtype=torch.bool)

my_tensor1, my_tensor2, my_tensor.dtype, my_tensor.type()
# (tensor([[0.],
#          [1.],
#          [2.]]),
#  tensor([[False, False, False, False],
#          [False, False,  True,  True],
#          [False, False, False,  True]]),
#  torch.bool,
#  'torch.BoolTensor')