Pytorch 2.1 基础知识大扫盲
Pytorch基础知识大扫盲
张量 Tensor
Tensor在TensorFlow中数据的核心单元就是Tensor。在PyTorch中Tensor负责存储基本数据，PyTorch也对Tensor提供了丰富的函数和方法，PyTorch中Tensor与Numpy的数组十分相似。并且PyTorch中定义的Tensor数据类型可以实现在GUPs上进行运算，并且只需要对变量做一些简单的类型转换就可以实现。
首先需要掌握Tensor定义不同数据类型的方法，常用的如下：
▷▷▷▷创建Tensor▷▷▷
Tensor的创建
torch.tensor(data, #功能解释：从data创建tensor #data:数据，可以是list，ndarray
dtype=None, #dtype：数据类型
device=None, #device：所在设备，cuda/cpu
requires_grad=False, #requires_grad：是否计算梯度
pin_memory=False) #pin_memory：是否存于锁页内存
Tensor类型转换
import torch ,numpy as np
# 类型转换
a = np.array([1,2,3,4,5])
before = id(a)
b = torch.from_numpy(a)
b[2] = 999
a,b,type(a),type(b)
>>>[ 1 2 999 4 5] tensor([ 1, 2, 999, 4, 5], dtype=torch.int32) <class 'numpy.ndarray'> <class 'torch.Tensor'>
在使用torch.from_numpy()这个函数的时候，传入的参数和返回的参数共享内存，所以当我修改b的时候，a也会被修改，但是ab的内存地址不一样
创建0、1组成的tensor
#创建0、1组成的tensor
torch.zeros(size, #size: 张量的形状，如(3, 3)、(3, 224,224)
out=None, #out : 输出的张量，将创建的tensor赋值给out，共享内存
dtype=None, #dtype : 数据类型
layout=torch.strided, #layout : 内存中布局形式，有 strided，sparse_coo(稀疏张量使用)等
device=None, #device : 所在设备，cuda/cpu
requires_grad=False)#requires_grad：是否需要计算梯度
#torch.zeros_like()
torch.zeros_like(input, #intput: 创建与input同形状的全0张量
dtype=None,
layout=None,
device=None,
requires_grad=False)
#torch.ones()
torch.ones(size,
out=None,
dtype=None,
layout=torch.strided,
device=None,
requires_grad=False
# torch.ones_like()
torch.ones_like(input,
dtype=None,
layout=None,
device=None,
requires_grad=False
创建其他数字的Tensor
# torch.full()
torch.full(size,
fill_value, #fill_value：张量中填充的值
out=None,
dtype=None,
layout=torch.strided,
device=None,
requires_grad=False
# torch.full_like()
torch.full_like(input,
fill_value,
dtype=None,
layout=None,
device=None,
requires_grad=False
e = torch.full((3,4), fill_value=3)
e
>>>tensor([[3, 3, 3, 3],
[3, 3, 3, 3],
[3, 3, 3, 3]])
标准阵的创建
# torch.eye()
'''torch.eye(n,
m=None,
out=None,
dtype=None,
layout=torch.strided,
device=None,
requires_grad=False)'''
f = torch.eye(n=3,m=4) #功能解释：创建n行m列的identify Matrix，注意：未输入m时，m=n
f
>>>tensor([[1., 0., 0., 0.],
[0., 1., 0., 0.],
[0., 0., 1., 0.]])
torch.empty()用法
# torch.empty()
g = torch.empty(2,3)
'''torch.empty(sizes,
out=None,
dtype=None,
layout=torch.strided,
device=None,
requires_grad=False)'''
# torch.empty_like()创建input.size()大小，值全为0的Tensor
'''torch.empty_like(input,
dtype=None,
layout=None,
device=None,
requires_grad=False)'''
g
>>>tensor([[ 1.6167e-30, 3.0816e-41, -4.8385e+02],
[ 4.5758e-41, 8.9683e-44, 0.0000e+00]])
创建序列数组
#torch.arange() 按照步长创建tensor
'''torch.arange(start=0,
end,
step=1,
out=None,
dtype=None,
layout=torch.strided,
device=None,
requires_grad=False)'''
#torch.range() 等差一维数组 同torch.arange
'''torch.range(start=0,
end,
step=1,
out=None,
dtype=None,
layout=torch.strided,
device=None,
requires_grad=False)'''
h = torch.arange(start=0,end=12,step=2)
h
>>>tensor([ 0, 2, 4, 6, 8, 10])
torch.linspace()
# torch.linspace() #步长(end-start)/steps等差的一维张量
'''torch.linspace(start,
end,
steps=100, #注意这里的steps和之前的step不一样，这里指的是一共要走100步
out=None,
dtype=None,
layout=torch.strided,
device=None,
requires_grad=False)'''
i = torch.linspace(start=0,end=11)
i
>>>tensor([ 0.0000, 0.1111, 0.2222, 0.3333, 0.4444, 0.5556, 0.6667, 0.7778,
0.8889, 1.0000, 1.1111, 1.2222, 1.3333, 1.4444, 1.5556, 1.6667,
1.7778, 1.8889, 2.0000, 2.1111, 2.2222, 2.3333, 2.4444, 2.5556,
2.6667, 2.7778, 2.8889, 3.0000, 3.1111, 3.2222, 3.3333, 3.4444,
3.5556, 3.6667, 3.7778, 3.8889, 4.0000, 4.1111, 4.2222, 4.3333,
4.4444, 4.5556, 4.6667, 4.7778, 4.8889, 5.0000, 5.1111, 5.2222,
5.3333, 5.4444, 5.5556, 5.6667, 5.7778, 5.8889, 6.0000, 6.1111,
6.2222, 6.3333, 6.4444, 6.5556, 6.6667, 6.7778, 6.8889, 7.0000,
7.1111, 7.2222, 7.3333, 7.4444, 7.5556, 7.6667, 7.7778, 7.8889,
8.0000, 8.1111, 8.2222, 8.3333, 8.4444, 8.5556, 8.6667, 8.7778,
8.8889, 9.0000, 9.1111, 9.2222, 9.3333, 9.4444, 9.5556, 9.6667,
9.7778, 9.8889, 10.0000, 10.1111, 10.2222, 10.3333, 10.4444, 10.5556,
10.6667, 10.7778, 10.8889, 11.0000])
torch.logspace()
#torch.logspace()
'''torch.logspace(start,
end,
steps=100,
out=None,
dtype=None,
layout=torch.strided,
device=None,
requires_grad=False)'''
j = torch.logspace(0,11)
j
>>>tensor([1.0000e+00, 1.2915e+00, 1.6681e+00, 2.1544e+00, 2.7826e+00, 3.5938e+00,
4.6416e+00, 5.9948e+00, 7.7426e+00, 1.0000e+01, 1.2915e+01, 1.6681e+01,
2.1544e+01, 2.7826e+01, 3.5938e+01, 4.6416e+01, 5.9948e+01, 7.7426e+01,
1.0000e+02, 1.2915e+02, 1.6681e+02, 2.1544e+02, 2.7826e+02, 3.5938e+02,
4.6416e+02, 5.9948e+02, 7.7426e+02, 1.0000e+03, 1.2915e+03, 1.6681e+03,
2.1544e+03, 2.7826e+03, 3.5938e+03, 4.6416e+03, 5.9948e+03, 7.7426e+03,
1.0000e+04, 1.2915e+04, 1.6681e+04, 2.1544e+04, 2.7826e+04, 3.5938e+04,
4.6416e+04, 5.9948e+04, 7.7426e+04, 1.0000e+05, 1.2915e+05, 1.6681e+05,
2.1544e+05, 2.7826e+05, 3.5938e+05, 4.6416e+05, 5.9948e+05, 7.7426e+05,
1.0000e+06, 1.2915e+06, 1.6681e+06, 2.1544e+06, 2.7826e+06, 3.5938e+06,
4.6416e+06, 5.9948e+06, 7.7426e+06, 1.0000e+07, 1.2915e+07, 1.6681e+07,
2.1544e+07, 2.7826e+07, 3.5938e+07, 4.6416e+07, 5.9948e+07, 7.7426e+07,
1.0000e+08, 1.2915e+08, 1.6681e+08, 2.1544e+08, 2.7826e+08, 3.5938e+08,
4.6416e+08, 5.9948e+08, 7.7426e+08, 1.0000e+09, 1.2915e+09, 1.6681e+09,
2.1544e+09, 2.7826e+09, 3.5938e+09, 4.6416e+09, 5.9948e+09, 7.7426e+09,
1.0000e+10, 1.2915e+10, 1.6681e+10, 2.1544e+10, 2.7826e+10, 3.5938e+10,
4.6416e+10, 5.9948e+10, 7.7426e+10, 1.0000e+11])
▷▷▷▷变形Tensor▷▷▷
拼接Tensor
#torch.cat()
'''torch.cat(seq,
dim=0,
out=None)'''
k = torch.arange(9).reshape(3,3)
l = torch.arange(9).reshape(3,3)
torch.cat((k,l),dim=0),torch.cat((k,l),dim=1) # 保留列，左右拼接 #保留行，上下拼接
>>>(tensor([[0, 1, 2],
[3, 4, 5],
[6, 7, 8],
[0, 1, 2],
[3, 4, 5],
[6, 7, 8]]),
tensor([[0, 1, 2, 0, 1, 2],
[3, 4, 5, 3, 4, 5],
[6, 7, 8, 6, 7, 8]]))
torch.squeeze() 去除那些维度大小为1的维度
#torch.squeeze()
'''torch.squeeze(input,
dim=None,
out=None)'''
torch.unbind() 去除某个维度
#torch.unbind()
'''torch.unbind(tensor,
dim=0)'''
torch.unsqueeze() 在指定位置添加维度
'''torch.unsqueeze(input,
dim,
out=None) '''
上一篇：Pytorch基础：Pytorch的优势
下一篇：Pytorch 学习笔记2 基础知识大扫盲
预备知识1.matplotlib的基本用法和常识
预备知识2.Pytorch下的线代
预备知识3.pandas的基本用法
posted on 2021-12-10 08:47 YangShusen' 阅读(463) 评论(0) 编辑 收藏 举报
