文章/答案/技术大牛

发布

社区首页 >专栏 >Deep Learning with PyTorch: A 60 Minute Blitz > Tensors

Deep Learning with PyTorch: A 60 Minute Blitz > Tensors

原创

望天

发布于 2024-06-06 13:36:31

20300

代码可运行

文章被收录于专栏：along的开发之旅along的开发之旅

运行总次数：0

代码可运行

主要是PyTorch官方原文，增加了Ascend npu的示例和实战。

Tensors

Tensors are a specialized data structure that are very similar to arrays and matrices. In PyTorch, we use tensors to encode the inputs and outputs of a model, as well as the model’s parameters.

Tensors are similar to NumPy’s ndarrays, except that tensors can run on GPUs or other specialized hardware to accelerate computing. If you’re familiar with ndarrays, you’ll be right at home with the Tensor API. If not, follow along in this quick API walkthrough.

import torch
import numpy as np

Tensor Initialization

Tensors can be initialized in various ways. Take a look at the following examples:

Directly from data

Tensors can be created directly from data. The data type is automatically inferred.

data = [[1, 2], [3, 4]]
x_data = torch.tensor(data)

From a NumPy array

Tensors can be created from NumPy arrays (and vice versa - see Bridge with NumPy).

np_array = np.array(data)
x_np = torch.from_numpy(np_array)

From another tensor:

The new tensor retains the properties (shape, datatype) of the argument tensor, unless explicitly overridden.

x_ones = torch.ones_like(x_data) # retains the properties of x_data
print(f"Ones Tensor: \n {x_ones} \n")

x_rand = torch.rand_like(x_data, dtype=torch.float) # overrides the datatype of x_data
print(f"Random Tensor: \n {x_rand} \n")

Ones Tensor: tensor([1, 1, 1, 1])
Random Tensor: tensor([0.8823, 0.9150, 0.3829, 0.9593])

With random or constant values:

shape is a tuple of tensor dimensions. In the functions below, it determines the dimensionality of the output tensor.

shape = (2, 3, )
rand_tensor = torch.rand(shape)
ones_tensor = torch.ones(shape)
zeros_tensor = torch.zeros(shape)

print(f"Random Tensor: \n {rand_tensor} \n")
print(f"Ones Tensor: \n {ones_tensor} \n")
print(f"Zeros Tensor: \n {zeros_tensor}")

Random Tensor:
 tensor([[0.3904, 0.6009, 0.2566],
        [0.7936, 0.9408, 0.1332]])

Ones Tensor:
 tensor([[1., 1., 1.],
        [1., 1., 1.]])

Zeros Tensor:
 tensor([[0., 0., 0.],
        [0., 0., 0.]])

Tensor Attributes

Tensor attributes describe their shape, datatype, and the device on which they are stored.

tensor = torch.rand(3, 4)

print(f"Shape of tensor: {tensor.shape}")
print(f"Datatype of tensor: {tensor.dtype}")
print(f"Device tensor is stored on: {tensor.device}")

Tensor Operations

Over 100 tensor operations, including transposing, indexing, slicing, mathematical operations, linear algebra, random sampling, and more are comprehensively described here.

Each of them can be run on the GPU (at typically higher speeds than on a CPU). If you’re using Colab, allocate a GPU by going to Edit > Notebook Settings.

# We move our tensor to the GPU if available
if torch.cuda.is_available():
  tensor = tensor.to('cuda')
  print(f"Device tensor is stored on: {tensor.device}")

Device tensor is stored on: cuda:0

Try out some of the operations from the list. If you’re familiar with the NumPy API, you’ll find the Tensor API a breeze to use.

Ascend NPU

如果在Ascend NPU上使用，简单可以使用from torch_npu.contrib import transfer_to_npu来自动迁移。

>>> torch.cuda.is_available()
False
>>> import torch_npu
>>> from torch_npu.contrib import transfer_to_npu
>>> torch.cuda.is_available()
True
>>> tensor = tensor.to('cuda')
>>> print(f"Device tensor is stored on: {tensor.device}")
Device tensor is stored on: npu:0

可以看到，当调用了transfer_to_npu后，cuda相关的调用被自动转到npu了，相关的判断也通过了。

或者直接使用npu相关的API，例如：

>>> import torch_npu
>>> torch.npu.is_available()
True
>>> tensor = tensor.npu()
>>> print(f"Device tensor is stored on: {tensor.device}")
Device tensor is stored on: npu:0

Standard numpy-like indexing and slicing:

tensor = torch.ones(4, 4)
tensor[:,1] = 0
print(tensor)

tensor([[1., 0., 1., 1.],
        [1., 0., 1., 1.],
        [1., 0., 1., 1.],
        [1., 0., 1., 1.]])

Joining tensors

You can use torch.cat to concatenate a sequence of tensors along a given dimension. See also torch.stack, another tensor joining op that is subtly different from torch.cat.

t1 = torch.cat([tensor, tensor, tensor], dim=1)
print(t1)

tensor([[1., 0., 1., 1., 1., 0., 1., 1., 1., 0., 1., 1.],
        [1., 0., 1., 1., 1., 0., 1., 1., 1., 0., 1., 1.],
        [1., 0., 1., 1., 1., 0., 1., 1., 1., 0., 1., 1.],
        [1., 0., 1., 1., 1., 0., 1., 1., 1., 0., 1., 1.]])

torch.cat 和 torch.stack 的dim都稍显复杂，可以通过官网case来了解。

torch.cat dim=0时，就是说在torch.shape的第0维度上进行cat连接，所以也就是行；dim=1时，就是在torch.shape的第1维度上进行cat拼接，所以也就是列。当维度更高时，可能就不好类比长宽高了，这时候根据torch.shape的维度来判断就好。

>>> x = torch.randn(2, 3)
>>> x
tensor([[ 0.0679, -0.3655, -1.5670],
        [-0.6854,  0.1267, -0.8296]])
>>> x0 = torch.cat((x, x, x), 0)
>>> x0.shape
torch.Size([6, 3])
>>> x0
tensor([[ 0.0679, -0.3655, -1.5670],
        [-0.6854,  0.1267, -0.8296],
        [ 0.0679, -0.3655, -1.5670],
        [-0.6854,  0.1267, -0.8296],
        [ 0.0679, -0.3655, -1.5670],
        [-0.6854,  0.1267, -0.8296]])        

>>> x1 = torch.cat((x, x, x), 1)
>>> x1.shape
torch.Size([2, 9])
>>> x1
tensor([[ 0.0679, -0.3655, -1.5670,  0.0679, -0.3655, -1.5670,  0.0679, -0.3655,
         -1.5670],
        [-0.6854,  0.1267, -0.8296, -0.6854,  0.1267, -0.8296, -0.6854,  0.1267,
         -0.8296]])

>>> x_0_t
tensor([[1.1100, 1.1200, 1.1300, 1.1400],
        [1.2100, 1.2200, 1.2300, 1.2400],
        [1.3100, 1.3200, 1.3300, 1.3400]])

>>> x_1_t
tensor([[2.1100, 2.1200, 2.1300, 2.1400],
        [2.2100, 2.2200, 2.2300, 2.2400],
        [2.3100, 2.3200, 2.3300, 2.3400]])

>>> y0 = torch.stack((x_0_t, x_1_t),  dim=0)
>>> y0.shape
torch.Size([2, 3, 4])
>>> y0
tensor([[[1.1100, 1.1200, 1.1300, 1.1400],
         [1.2100, 1.2200, 1.2300, 1.2400],
         [1.3100, 1.3200, 1.3300, 1.3400]],

        [[2.1100, 2.1200, 2.1300, 2.1400],
         [2.2100, 2.2200, 2.2300, 2.2400],
         [2.3100, 2.3200, 2.3300, 2.3400]]])

>>> y1 = torch.stack((x_0_t, x_1_t),  dim=1)
>>> y1.shape
torch.Size([3, 2, 4])
>>> y1
tensor([[[1.1100, 1.1200, 1.1300, 1.1400],
         [2.1100, 2.1200, 2.1300, 2.1400]],

        [[1.2100, 1.2200, 1.2300, 1.2400],
         [2.2100, 2.2200, 2.2300, 2.2400]],

        [[1.3100, 1.3200, 1.3300, 1.3400],
         [2.3100, 2.3200, 2.3300, 2.3400]]])
>>> y2 = torch.stack((x_0_t, x_1_t),  dim=2)
>>> y2.shape
torch.Size([3, 4, 2])
>>> y2
tensor([[[1.1100, 2.1100],
         [1.1200, 2.1200],
         [1.1300, 2.1300],
         [1.1400, 2.1400]],

        [[1.2100, 2.2100],
         [1.2200, 2.2200],
         [1.2300, 2.2300],
         [1.2400, 2.2400]],

        [[1.3100, 2.3100],
         [1.3200, 2.3200],
         [1.3300, 2.3300],
         [1.3400, 2.3400]]])

Multiplying tensors

tensor = torch.ones(4, 4)
tensor[:,1] = 0
print(tensor)
# This computes the element-wise product
print(f"tensor.mul(tensor) \n {tensor.mul(tensor)} \n")
# Alternative syntax:
print(f"tensor * tensor \n {tensor * tensor}")

tensor.mul(tensor)
 tensor([[1., 0., 1., 1.],
        [1., 0., 1., 1.],
        [1., 0., 1., 1.],
        [1., 0., 1., 1.]])

tensor * tensor
 tensor([[1., 0., 1., 1.],
        [1., 0., 1., 1.],
        [1., 0., 1., 1.],
        [1., 0., 1., 1.]])

This computes the matrix multiplication between two tensors

print(f"tensor.matmul(tensor.T) \n {tensor.matmul(tensor.T)} \n")
# Alternative syntax:
print(f"tensor @ tensor.T \n {tensor @ tensor.T}")

tensor.matmul(tensor.T)
 tensor([[3., 3., 3., 3.],
        [3., 3., 3., 3.],
        [3., 3., 3., 3.],
        [3., 3., 3., 3.]])

tensor @ tensor.T
 tensor([[3., 3., 3., 3.],
        [3., 3., 3., 3.],
        [3., 3., 3., 3.],
        [3., 3., 3., 3.]])

In-place operations

Operations that have a suffix are in-place. For example: x.copy(y), x.t_(), will change x.

print(tensor, "\n")
tensor.add_(5)
print(tensor)

tensor([[1., 0., 1., 1.],
        [1., 0., 1., 1.],
        [1., 0., 1., 1.],
        [1., 0., 1., 1.]])

tensor([[6., 5., 6., 6.],
        [6., 5., 6., 6.],
        [6., 5., 6., 6.],
        [6., 5., 6., 6.]])

In-place operations save some memory, but can be problematic when computing derivatives because of an immediate loss of history. Hence, their use is discouraged.

Bridge with NumPy

Tensors on the CPU and NumPy arrays can share their underlying memory locations, and changing one will change the other.

Tensor to NumPy array

t = torch.ones(5)
print(f"t: {t}")
n = t.numpy()
print(f"n: {n}")

t: tensor([1., 1., 1., 1., 1.])
n: [1. 1. 1. 1. 1.]

A change in the tensor reflects in the NumPy array.

t.add_(1)
print(f"t: {t}")
print(f"n: {n}")

t: tensor([2., 2., 2., 2., 2.])
n: [2. 2. 2. 2. 2.]

但是GPU上的数据，修改是不会联动的。

>>> t = torch.ones(5).npu()
>>> t
tensor([1., 1., 1., 1., 1.], device='npu:0')
>>> n = t.cpu().numpy()
>>> n
array([1., 1., 1., 1., 1.], dtype=float32)
>>> t.add_(1)
tensor([2., 2., 2., 2., 2.], device='npu:0')
>>> t
tensor([2., 2., 2., 2., 2.], device='npu:0')
>>> n
array([1., 1., 1., 1., 1.], dtype=float32)

NumPy array to Tensor

n = np.ones(5)
t = torch.from_numpy(n)

Changes in the NumPy array reflects in the tensor.

np.add(n, 1, out=n)
print(f"t: {t}")
print(f"n: {n}")

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

同样不会影响到GPU上的数据，因为GPU的数据和CPU是分离的，需要先从GPUcopy到CPU上。

>>> import numpy as np
>>> n = np.ones(5)
>>> t = torch.from_numpy(n).npu()
>>> n
array([1., 1., 1., 1., 1.])
>>> t
tensor([1., 1., 1., 1., 1.], device='npu:0', dtype=torch.float64)
>>> np.add(n, 1, out=n)
array([2., 2., 2., 2., 2.])
>>> n
array([2., 2., 2., 2., 2.])
>>> t
tensor([1., 1., 1., 1., 1.], device='npu:0', dtype=torch.float64)