Pytorch for former Torch users - Tensors

Tensors

Pytorch Medium Tensor and Torch Medium Tensor Have almost the same usage
Create a (5*7) Uninitialized... Of size tensor

import torch
a = torch.empty(5, 7, dtype=torch.floar)

Initializes a double precision floating-point number randomly obtained from the standard normal distribution tensor

b = torch.randn(5, 7, dtype=torch.double)
print(b.size())
# torch.Size([5, 7])

Note: torch.Size It's actually a tuple (tuple)

Inplace / Out-of-place

tensor All local operations on the are suffixed _
for example ,add It's an ex situ version , add_ It is an in place version

Some operations , for example narrow There is no in place version , And like fill_ There is no easy version for the operation of

Zero Indexing

Pytorch in tensor The index is from 0 At the beginning
a = b[0, 3] # from b Select the first 1 That's ok , The first 4 Column
Tensors You can also use Python Slice index for
b = a[:, 3:5] # Select all lines , The first 4 And the 5 Column

No Camel Casing Humpless spelling

All function names are no longer spelled hump . for example indexAdd Now rewrite to index_add_

x = torch.ones(5, 5);
# tensor([[ 1., 1., 1., 1., 1.],
# [ 1., 1., 1., 1., 1.],
# [ 1., 1., 1., 1., 1.],
# [ 1., 1., 1., 1., 1.],
# [ 1., 1., 1., 1., 1.]])
z = torch.empty(5, 2)
z[:, 0] = 10
z[:, 1] = 100
# tensor([[ 10., 100.],
# [ 10., 100.],
# [ 10., 100.],
# [ 10., 100.],
# [ 10., 100.]])
x.index_add_(1, torch.tensor([4, 0]), z)
#  stay x The first dimensional direction of ( horizontal direction ), x The fifth column of ( The index for 4) Add z The first column ,x The first column ( The index for 0) Add z The second column of 
# tensor([[ 101., 1., 1., 1., 11.],
# [ 101., 1., 1., 1., 11.],
# [ 101., 1., 1., 1., 11.],
# [ 101., 1., 1., 1., 11.],
# [ 101., 1., 1., 1., 11.]])

Numpy Bridge

Numpy array and torch Tensor The transition between them is very simple , Memory addresses are shared between the two , That is, changing one will change the other .

hold torch Tensor convert to numpy Array

a = torch.ones(5)
# tensor([ 1., 1., 1., 1., 1.])
b = a.numpy()
# [1. 1. 1. 1. 1.]
a.add_(1)
# tensor([ 2., 2., 2., 2., 2.])
# [2. 2. 2. 2. 2.]

hold numpy Array convert to torch Tensor

a = np.ones(5)
b = torch.from_numpy(a)
# [1. 1. 1. 1. 1.] 
# tensor([ 1., 1., 1., 1., 1.], dtype=torch.float64)

CUDA Tensors

stay Pytorch in CUDA Tensors Simple to use , from CPU Change one CUDA Tensors To GPU Will save its original type

# Let us run this cell only cuda is available
if torch.cuda.is_available():
    # create a LongTensor and transfer it
    # to GPU as torch.cuda.LongTensor
    a = torch.full((10, ), 2, device=torch.device("cuda"))
    print(a)
    b = a.to(torch.device("cpu"))
    # transfer it to CPU, back to
    # being a torch.LongTensor