pytorch基础（1）

基本数据类型和tensor

 

import torch
import numpy as np

#array 和 tensor的转换
array = np.array([1.1,2,3])
tensorArray = torch.from_numpy(array) #array对象变为tensor对象
array1  = tensorArray.numpy()#tensor对象变为array对象
print(array,'\t', tensorArray, '\t', array1 )

#torch拥有和numpy一样的处理数据的能力
print(torch.sin(tensorArray))
print(np.ones([2,5]))#两行五列
print(np.ones(2))#一行两个数字
a = torch.randn(2, 3)#两行三列的正态分布
print(a)
print(a.size(0),a.size(1),a.shape[1])#2,3,3   0代表行，1代表对应的列数
print(a.shape)#torch.Size([2,3])
print(a.type())#torch.FloatTensor
isinstance(a, torch.DoubleTensor)#false
isinstance(a, torch.FloatTensor)#true
a1 = a.cuda()
print(isinstance(a1,torch.FloatTensor))#false
print(isinstance(a1,torch.cuda.FloatTensor))#true，#torch里面的数据不同于torch.cuda里面的数据

#torch的tensor对象
tensor1 = torch.tensor(1)
print(tensor1)#tensor(1)
tensor2 = torch.tensor(1.2)
print(tensor2)#tensor(1.2000)
print(tensor2.shape)#torch.Size([])
print(len(tensor2.shape))#0,当tensor只是一个数字的时候，他的维度是0，所以他的size是[],shape为0
tensor3 = torch.tensor([1.1])#一维的列表，所以输出维度是1
print(tensor3,tensor3.shape)#tensor([1.1000]) torch.Size([1])
tensor4 = torch.FloatTensor(1)#注意此时1代表随机返回一个FloatTensor对象
print(tensor4)#tensor([1.1000])
tensor5 = torch.FloatTensor(3)#考虑一下tensor和FloatTensor的差别
print(tensor5)#tensor([0.0000e+00, 0.0000e+00, 6.8645e+36])

 

切片

 

import torch
import numpy as np

#tensor和随机数
a = torch.rand(2, 3, 28, 28)
print(a, a.shape)#随机生成一个2*3*28*28的四维矩阵（可以看成声明一个四维矩阵），torch.Size([2, 3, 28, 28])
                  #四维适合做CNN ,三维适合RNN，二维适合batch
print(a.numel())#4707,计算元素个数
print(a.dim())#4
print(torch.tensor(1).dim())#0，
print(torch.empty(1))#一维数字0，tensor([0.])
print(torch.Tensor(2,3).type())#默认是torch.FloatTensor
print(torch.IntTensor(2,3))#2*3
print(torch.tensor([1,1]).type())#torch.LongTensor
print(torch.tensor([1.2,1]).type())#torch.FloatTensor
print(torch.rand(3,3))#取值范围为0到1之间
print(torch.rand_like(torch.rand(3,3)))#rand_like直接继承了参数的行和列，生成3*3的0到1之间的随机矩阵
print(torch.randint(1,10,(3,3)))#取值在1到10之间（左闭右开）大小为3*3的矩阵
print(torch.randn(3,3))#3*3矩阵，服从均值为0，方差为1的正态分布
print(torch.normal(mean=torch.full([10],0),std = torch.arange(1,0,-0.1)))#均值为0方差递减的10*1一维矩阵
print(torch.full([2,3],7))#2*3全为7的二维矩阵
print(torch.full([],7))#数字7维度0
print([1],7)#一维1*1矩阵元素为7
print(torch.logspace(0,1,steps=10))#log(10^0)到log(10^1)中间取10个数

#切片
a = torch.rand(4,3,28,28)
print(a[0].shape)#torch.Size([3,28,28])
print(a[0,0].shape)#torch.Size([28, 28])
print(a[0,0,2,4])#tensor(0.6186)
print(a[:2].shape)#torch.Size([2, 3, 28, 28])
print(a[:2,1:,:,:].shape)#torch.Size([2, 2, 28, 28])
print(a[:2,-1:,:,:].shape)#torch.Size([2, 1, 28, 28])

print(a[:,:,0:28:2,0:28:2].shape)#torch.Size([4, 3, 14, 14])
print(a[:,:,::2,::2].shape)#torch.Size([4, 3, 14, 14])
print(a.index_select(1,torch.arange(1)).shape)#torch.Size([4, 1, 28, 28])

x = torch.randn(3,4)
mask = x.ge(0.5)#比0.5大的标记为true
print(mask)
torch.masked_select(x,mask)#把为true的选择出来
torch.masked_select(x,mask).shape

src =torch.tensor([[4,3,5],[6,7,8]])
taa = torch.take(src, torch.tensor([0,3,5]))#展平后按照位置选数据
print(taa)

 

 

维度变换

 

import torch
import numpy as np
#维度变化
#View reshape
a = torch.rand(4,1,28,28)#四张图片，通道数是1，长宽是28*28
print(a.shape)#torch.Size([4, 1, 28, 28])
print(a.view(4,1*28*28).shape)#torch.Size([4, 784]),把后三维展成一行
print(a.view(4*28,28).shape)#torch.Size([112, 28])变成112行28列的二维数据
print(a.view(4*1,28,28).shape)#torch.Size([4, 28, 28])要理解对应的图片的物理意义
b = a.view(4,784)
print(b.view(4,28,28,1).shape)#torch.Size([4, 28, 28, 1]),b变成的数据不是a(一定要注意)
#print(a.view(4,783))#尺寸不一致会报错

#unsqueeze,增加维度，但不会影响数据的变化
#数据的范围是[-a.dim()-1,a.dim()+1)
print()#下面例子是[-5,5)
print(a.unsqueeze(0).shape)#torch.Size([1, 4, 1, 28, 28])
print(a.unsqueeze(-1).shape)#torch.Size([4, 1, 28, 28, 1])
print(a.unsqueeze(4).shape)#torch.Size([4, 1, 28, 28, 1])
print(a.unsqueeze(-4).shape)#torch.Size([4, 1, 1, 28, 28])
print(a.unsqueeze(-5).shape)#torch.Size([1, 4, 1, 28, 28])
#print(a.unsqueeze(5).shape)
a = torch.tensor([1.2,2.3])#a的shape是[2]
print(a.unsqueeze(-1))#tensor([[1.2000],
                        #[2.3000]])变成2行一列
print(a.unsqueeze(0))#tensor([[1.2000, 2.3000]])#shape变成[1,2],即一行二列
b = torch.rand(32)
f = torch.rand(4,3,14,14)
b = b.unsqueeze(1).unsqueeze(2).unsqueeze(0)#torch.Size([1, 32, 1, 1])
print(b.shape)

#维度减少
print()
print(b.shape)#torch.Size([1, 32, 1, 1])
print(b.squeeze().shape)#torch.Size([32]),所有为1的被挤压
print(b.squeeze(-1).shape)#torch.Size([1, 32, 1])
print(b.squeeze(0).shape)#torch.Size([32, 1, 1])
print(b.squeeze(1).shape)#torch.Size([1, 32, 1, 1]),因为不等于1 所以没有被挤压
print(b.squeeze(-4).shape)#torch.Size([32, 1, 1])

#expand扩展数据，进行数据拷贝，但不会主动复制数据，只会在需要的时候复制，推荐使用
print()
print(b.shape)#torch.Size([1, 32, 1, 1])
print(b.expand(4,32,14,14).shape)#torch.Size([4, 32, 14, 14]),只能对维度是1 的进行扩展
print(b.expand(-1,32,-1,-1).shape)#torch.Size([1, 32, 1, 1]),其他维度为-1，这样可以进行原维度不是一的进行扩展同样大小的维度
print(b.expand(-1,32,-1,-4).shape)#torch.Size([1, 32, 1, -4]) -4是无意义的

#repeat表示在原来维度上拷贝多少次，而不是扩展到多少，这个方法申请了新的空间，对空间使用加大
print()
print(b.shape)#torch.Size([1, 32, 1, 1])
print(b.repeat(4,32,1,1).shape)#torch.Size([4, 1024, 1, 1]),第二维表示拷贝愿来的32倍
print(b.repeat(4,1,1,1).shape)#torch.Size([4, 32, 1, 1])
print(b.repeat(4,1,32,32).shape)#torch.Size([4, 32, 32, 32])

#transpose实现指定维度之间的交换
a = torch.rand(4,3,32,32)
print(a.shape)#torch.Size([4, 3, 32, 32])
a1 = a.transpose(1,3).contiguous().view(4,3*32*32).view(4,32,32,3).transpose(1,3)
print(a1.shape)#torch.Size([4, 3, 32, 32])
print(torch.all(torch.eq(a,a1)))#tensor(True)

#premute实现指定维度位置交换到指定位置
print(a.permute(0,2,3,1).shape)#torch.Size([4, 32, 32, 3])