Shulou(Shulou.com)06/03 Report--

This article mainly introduces how to use Tensor of pytorch. It is very detailed and has certain reference value. Friends who are interested must finish reading it.

1 、 Tensors

Tensors are similar to NumPy's ndaeeays, the difference is that it can be used and accelerated on GPU.

Import package

From _ _ future__ import print_functionimport torch establishes a matrix of 5x _ 3, uninitialized x = torch.empty (5prime3) print (x)

Out

Tensor ([[1.4395e-36, 4.5848e-41, 1.4395e-36], [4.5848e-41, 1.4395e-36, 4.5848e-41], [1.4395e-36, 4.5848e-41, 2.8026e-45], [- 1.9501e+00, 8.5165e+23, 0.0000e+00], [2.5223e-43, 0.0000e+00] 0.0000e+00]]) establish a random initialization matrix x = torch.rand (5) print (x)

Out

Tensor ([[0.8074, 0.9175, 0.8109], [0.3313, 0.5902, 0.9179], [0.6562, 0.3283, 0.9798], [0.8218, 0.0817,0.4454], [0.5934,0.0040,0.3411]) to establish zero initialization matrix The data type is Long...x = torch.zeros (5dtype = torch.long) print (x).

Out

Tensor ([[0,0,0], [0,0,0], [0,0,0], [0,0,0], [0,0,0]]) establishes a tensor data from datax = torch.tensor ([5.5jue 3]) print (x).

Out

Tensor ([5.5000, 3.0000])

A new tensor is formed on the basis of the original tnesor, which will inherit the shapee and dtype attributes of the original tensor. Of course, we can also modify these attributes.

X = x.new_ones (5maiden dtype = torch.double) print (x) x = torch.randn_like (xMagne Dype = torch.float) print (x)

Out

Tensor ([1, 1, 1.], [1, 1, 1.], [1, 1, 1.], [1, 1, 1.], [1, 1, 1.], dtype=torch.float64) tensor ([[- 0.0730,-0.0716,-0.8259], [- 1.7004, 0.8790] -0.0659], [- 0.8969, 0.8736,-0.6035], [- 0.1539,-2.9178,-0.7456], [- 0.0245, 0.4075, 1.4904]]) get the sizeprint of tensor (x.size ())

Out

Torch.Size ([5,3])

Torch.size is a tuple that supports all tuple operations.

2. Four ways to add the operation of Tensor

Method 1 L

Print (xroomy)

Method two

Print (torch.add (XBY))

Method 3: output to additional tensor

Result = torch.empty (5 out= result 3) torch.add (XJI y, out= result) print (result)

Method 4: replace in place-the result is stored in y

Print (y) all in-place replacement

All operations that replace tensor in place have a suffix, such as x.copy (y), which changes x

Use standard numpy to operate print (x [: 1])

Out

Tensor ([- 0.0716, 0.8790, 0.8736,-2.9178, 0.4075]) uses torch.view to change the shape of tensor x = torch.randn (4je 4) y = x.view (16) z = x.view (- 1pm 8) # the size-1 is inferred from other dimensionsprint (x.size (), y.xize (), z.size ())

Out

Torch.Size ([4,4]) torch.Size ([16]) torch.Size ([2,8]) tensor converted to numpy, using itemx = torch.rnadn (1) print (x) print (x.item ()) Torch Tensor and numpy conversion a = torch.ones (5) print (a)

Out

Tensor ([1, 1, 1, 1.])

And changing the value of tensor will also change the value of numpy.

A.add1 print (a) print (b)

Out

Tensor ([2, 2, 2, 2.]) [2. 2. 2. 2.] Transform numpy array into pytorch Tensorimport numpy as npa = np.ones (5) b = torch.from_numpy (a) np.add (a) print (a) print (b)

Out

[2. 2. 2. 2.] tensor ([2, 2, 2, 2.], dtype=torch.float64)

All tensor on cpu supports numpy conversion, except char-shaped tensor

CUDA Tensors

Tensors can be moved to other devices using the .to method

... if torch.cuda.is_avaulable (): device = torch.device ("cuda") y = torch.ones_like (x _ my device = devcie) x = x.to (device) z = x _ ray print (z) print (z.to ("cpu", torch.double)).

Out

Tensor ([- 1.0620], device='cuda:0') tensor ([- 1.0620], dtype=torch.float64) these are all the contents of this article entitled "how to use the Tensor of pytorch". Thank you for reading! Hope to share the content to help you, more related knowledge, welcome to follow the industry information channel!

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.