import torch

 import numpy as np

 #array 和 tensor的转换

 array = np.array([1.1,,])

 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([,]))#两行五列

 print(np.ones())#一行两个数字

 a = torch.randn(, )#两行三列的正态分布

 print(a)

 print(a.size(),a.size(),a.shape[])#,,   0代表行，1代表对应的列数

 print(a.shape)#torch.Size([,])

 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()

 print(tensor1)#tensor()

 tensor2 = torch.tensor(1.2)

 print(tensor2)#tensor(1.2000)

 print(tensor2.shape)#torch.Size([])

 print(len(tensor2.shape))#,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([])

 tensor4 = torch.FloatTensor()#注意此时1代表随机返回一个FloatTensor对象

 print(tensor4)#tensor([1.1000])

 tensor5 = torch.FloatTensor()#考虑一下tensor和FloatTensor的差别

 print(tensor5)#tensor([0.0000e+00, 0.0000e+00, 6.8645e+36])

 import torch

 import numpy as np

 #tensor和随机数

 a = torch.rand(, , , )

 print(a, a.shape)#随机生成一个2***28的四维矩阵（可以看成声明一个四维矩阵），torch.Size([, , , ])

                   #四维适合做CNN ,三维适合RNN，二维适合batch

 print(a.numel())#,计算元素个数

 print(a.dim())#

 print(torch.tensor().dim())#，

 print(torch.empty())#一维数字0，tensor([.])

 print(torch.Tensor(,).type())#默认是torch.FloatTensor

 print(torch.IntTensor(,))#*

 print(torch.tensor([,]).type())#torch.LongTensor

 print(torch.tensor([1.2,]).type())#torch.FloatTensor

 print(torch.rand(,))#取值范围为0到1之间

 print(torch.rand_like(torch.rand(,)))#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(,,(,)))#取值在1到10之间（左闭右开）大小为3*3的矩阵

 print(torch.randn(,))#*3矩阵，服从均值为0，方差为1的正态分布

 print(torch.normal(mean=torch.full([],),std = torch.arange(,,-0.1)))#均值为0方差递减的10*1一维矩阵

 print(torch.full([,],))#*3全为7的二维矩阵

 print(torch.full([],))#数字7维度0

 print([],)#一维1*1矩阵元素为7

 print(torch.logspace(,,steps=))#log(^)到log(^)中间取10个数

 #切片

 a = torch.rand(,,,)

 print(a[].shape)#torch.Size([,,])

 print(a[,].shape)#torch.Size([, ])

 print(a[,,,])#tensor(0.6186)

 print(a[:].shape)#torch.Size([, , , ])

 print(a[:,:,:,:].shape)#torch.Size([, , , ])

 print(a[:,-:,:,:].shape)#torch.Size([, , , ])

 print(a[:,:,::,::].shape)#torch.Size([, , , ])

 print(a[:,:,::,::].shape)#torch.Size([, , , ])

 print(a.index_select(,torch.arange()).shape)#torch.Size([, , , ])

 x = torch.randn(,)

 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([[,,],[,,]])

 taa = torch.take(src, torch.tensor([,,]))#展平后按照位置选数据

 print(taa)

 import torch

 import numpy as np

 #维度变化

 #View reshape

 a = torch.rand(,,,)#四张图片，通道数是1，长宽是28*

 print(a.shape)#torch.Size([, , , ])

 print(a.view(,**).shape)#torch.Size([, ]),把后三维展成一行

 print(a.view(*,).shape)#torch.Size([, ])变成112行28列的二维数据

 print(a.view(*,,).shape)#torch.Size([, , ])要理解对应的图片的物理意义

 b = a.view(,)

 print(b.view(,,,).shape)#torch.Size([, , , ]),b变成的数据不是a(一定要注意)

 #print(a.view(,))#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()-,a.dim()+)

 print()#下面例子是[-,)

 print(a.unsqueeze().shape)#torch.Size([, , , , ])

 print(a.unsqueeze(-).shape)#torch.Size([, , , , ])

 print(a.unsqueeze().shape)#torch.Size([, , , , ])

 print(a.unsqueeze(-).shape)#torch.Size([, , , , ])

 print(a.unsqueeze(-).shape)#torch.Size([, , , , ])

 #print(a.unsqueeze().shape)

 a = torch.tensor([1.2,2.3])#a的shape是[]

 print(a.unsqueeze(-))#tensor([[1.2000],

                         #[2.3000]])变成2行一列

 print(a.unsqueeze())#tensor([[1.2000, 2.3000]])#shape变成[,],即一行二列

 b = torch.rand()

 f = torch.rand(,,,)

 b = b.unsqueeze().unsqueeze().unsqueeze()#torch.Size([, , , ])

 print(b.shape)

 #维度减少

 print()

 print(b.shape)#torch.Size([, , , ])

 print(b.squeeze().shape)#torch.Size([]),所有为1的被挤压

 print(b.squeeze(-).shape)#torch.Size([, , ])

 print(b.squeeze().shape)#torch.Size([, , ])

 print(b.squeeze().shape)#torch.Size([, , , ]),因为不等于1 所以没有被挤压

 print(b.squeeze(-).shape)#torch.Size([, , ])

 #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([, , , ])

 print(b.expand(,,,).shape)#torch.Size([, , , ]),只能对维度是1 的进行扩展

 print(b.expand(-,,-,-).shape)#torch.Size([, , , ]),其他维度为-，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(-,,-,-).shape)#torch.Size([, , , -]) -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([, , , ])

 print(b.repeat(,,,).shape)#torch.Size([, , , ]),第二维表示拷贝愿来的32倍

 print(b.repeat(,,,).shape)#torch.Size([, , , ])

 print(b.repeat(,,,).shape)#torch.Size([, , , ])

 #transpose实现指定维度之间的交换

 a = torch.rand(,,,)

 print(a.shape)#torch.Size([, , , ])

 a1 = a.transpose(,).contiguous().view(,**).view(,,,).transpose(,)

 print(a1.shape)#torch.Size([, , , ])

 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(,,,).shape)#torch.Size([, , , ])