PyTorch基础

红圈圈里的数据类型比较重要

瞎贴一波：

一、数据类型

a = torch.randn(2,3)
print(a)
print(type(a))
print(isinstance(a, torch.FloatTensor))#合法化检验
print(isinstance(a, torch.cuda.FloatTensor))#合法化检验

a = a.cuda()#返回一个GPU上的引用
print(isinstance(a, torch.cuda.FloatTensor))#合法化检验

>>>
tensor([[ 1.0300, -0.7234, -0.5679],
        [-0.7928, -0.2508, -0.9085]])
<class 'torch.Tensor'>
True
False
True

torch.tensor(1.)

>>>tensor(1.)

torch.tensor(1.3)

>>>tensor(1.3000)


torch.tensor([1.])
#torch.tensor([1.])

>>>tensor([1.])


a = torch.tensor(2.2)
a.shape

>>>torch.Size([])

len(a.shape)

>>>0

a.size()

>>>torch.Size([])

torch.tensor([1.1])

>>>tensor([1.1000])

torch.tensor([1.1,2.2])

>>>tensor([1.1000, 2.2000])

torch.FloatTensor(1)

>>>tensor([1.4574e-43])

torch.FloatTensor(2)

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

data = np.ones(2)
data

>>>array([1., 1.])

torch.from_numpy(data)

>>>tensor([1., 1.], dtype=torch.float64)

a = torch.ones(2)
a.shape

>>>torch.Size([2])

a = torch.randn(2,3)
a

>>>tensor([[ 0.5736,  0.4078, -0.8879],
        [-0.0126,  0.4756, -0.0214]])


a.shape

>>>torch.Size([2, 3])

a.size(0)

>>>2

a.size(1)

>>>3

a = torch.rand(2,1,4,4)
a

>>>tensor([[[[0.4224, 0.5601, 0.3603, 0.7440],
          [0.2749, 0.4182, 0.5837, 0.1956],
          [0.9536, 0.9231, 0.6720, 0.4501],
          [0.0255, 0.0731, 0.7247, 0.3907]]],


        [[[0.0651, 0.7923, 0.1018, 0.3250],
          [0.7650, 0.5583, 0.5320, 0.9807],
          [0.2130, 0.2525, 0.7932, 0.0258],
          [0.7981, 0.6380, 0.1390, 0.2147]]]])


a.numel() #元素个数

>>>32

a.dim()  #维度

>>>4

二、创建Tensor

1.import from numpy，从numpy数据中载入

a = np.array([2,3.3])
torch.from_numpy(a)

>>>tensor([2.0000, 3.3000], dtype=torch.float64)

a = np.ones([2,3])
torch.from_numpy(a)

>>>tensor([[1., 1., 1.],
        [1., 1., 1.]], dtype=torch.float64)

2.import from List，从列表类型数据中载入

torch.tensor([2., 3.2]) #直接接受数据，经常使用，注意是小写
>>>tensor([2.0000, 3.2000])

torch.FloatTensor(4,3)#设定维度，经常使用
>>>tensor([[0., 0., 0.],
        [0., 0., 0.],
        [0., 0., 0.],
        [0., 0., 0.]])


torch.FloatTensor([2.,3.2]) #不轻易用，容易和小写的tensor混淆
>>>tensor([2.0000, 3.2000])

torch.IntTensor(2,3) #未初始化，数据分布差异太大
>>>tensor([[0, 0, 1],
        [0, 1, 0]], dtype=torch.int32)

torch.FloatTensor(2,3) #未初始化
>>>tensor([[7.4715e+37, 6.6001e-43, 8.4078e-45],
        [0.0000e+00, 1.4013e-45, 0.0000e+00]])


torch.tensor([1.2, 3]).type()
>>>'torch.FloatTensor'    #默认情况是FloatTensor

torch.set_default_tensor_type(torch.DoubleTensor)     #修改默认情况
torch.tensor([1.2, 3]).type()
>>>'torch.DoubleTensor' 

 3. rand/rand_like，randint

torch.rand(3,3) #采样(0,1)之间的分布
>>>tensor([[0.2816, 0.4244, 0.1964],
        [0.5480, 0.6377, 0.2992],
        [0.4763, 0.4480, 0.1515]])

a = torch.rand(3,3) 
torch.rand_like(a)#读出a的维度后再送到函数里
>>>tensor([[0.4966, 0.2000, 0.9064],
        [0.8584, 0.1375, 0.7557],
        [0.2186, 0.9776, 0.8252]])

torch.randint(1,10,[3,3]) #(min,max,[size[0],size[1]])
>>>tensor([[2, 1, 7],
        [7, 7, 6],
        [8, 5, 3]])

###正态分布的采样,均值为0，方差为1
torch.randn(3,3)
>>>tensor([[-0.3037,  1.2203, -0.2857],
        [ 0.4289,  0.3293,  0.6834],
        [-0.5883,  0.6679, -0.1545]])

torch.normal(mean=torch.full([10],0), std=torch.arange(1, 0, -0.1))
>>>tensor([-0.1700, -1.2166,  0.0035, -0.4357, -0.0571, -0.9798, -0.1286,  0.1009,
         0.2687, -0.1457])

torch.full([2,3],7)
>>>'torch.DoubleTensor' #之前修改了默认数据类型

torch.full([],7) #表示生成标量
>>>'torch.DoubleTensor'

torch.full([1],7) #生成维度为1的vector
>>>tensor([7.])

torch.arange(0,10)
>>>tensor([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])

torch.arange(0,10,2)
>>>tensor([0, 2, 4, 6, 8])

torch.linspace(0,10,steps=4)
>>>tensor([ 0.0000,  3.3333,  6.6667, 10.0000])

torch.linspace(0,10,steps=10)
>>>tensor([ 0.0000,  1.1111,  2.2222,  3.3333,  4.4444,  5.5556,  6.6667,  7.7778,
         8.8889, 10.0000])

torch.linspace(0,10,steps=11)
>>>tensor([ 0.,  1.,  2.,  3.,  4.,  5.,  6.,  7.,  8.,  9., 10.])

torch.logspace(0,-1,steps=10) #10^0,...,10^(-1)
>>>tensor([1.0000, 0.7743, 0.5995, 0.4642, 0.3594, 0.2783, 0.2154, 0.1668, 0.1292,
        0.1000])

torch.logspace(0,1,steps=10) #10^0,...,10^1
>>>tensor([ 1.0000,  1.2915,  1.6681,  2.1544,  2.7826,  3.5938,  4.6416,  5.9948,
         7.7426, 10.0000])

torch.ones(3,3)
>>>tensor([[1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.]])

torch.zeros(3,3)
>>>tensor([[0., 0., 0.],
        [0., 0., 0.],
        [0., 0., 0.]])

torch.eye(3,4)
>>>tensor([[1., 0., 0., 0.],
        [0., 1., 0., 0.],
        [0., 0., 1., 0.]])

a = torch.zeros(3,3)
torch.ones_like(a) #读取tensor维度再生成
>>>tensor([[1., 1., 1.],
        [1., 1., 1.],
        [1., 1., 1.]])

4. shuffle方法

a = torch.rand(2,3)
b = torch.rand(2,2)
a,b

>>>(tensor([[0.3372, 0.5688, 0.9331],
         [0.6488, 0.4826, 0.6608]]), tensor([[0.4330, 0.8487],
         [0.0610, 0.5291]]))

idx = torch.randperm(2)
idx

>>>tensor([1, 0])  #表示经过了一次shuffle，从0,1->1,0

a[idx]
>>>tensor([[0.6488, 0.4826, 0.6608],
        [0.3372, 0.5688, 0.9331]])

b[idx]
>>>tensor([[0.0610, 0.5291],
        [0.4330, 0.8487]])

三、索引与切片

1、给定索引号

a = torch.rand(4,3,28,28)
a[0].shape #第0张图
>>>torch.Size([3, 28, 28])

a[0,0].shape #第0张图，第0个通道
>>>torch.Size([28, 28])

a[0,0,2,4] #第0张图，第0个通道，第2行第4列的像素点
>>>tensor(0.7300)

a.shape
>>>torch.Size([4, 3, 28, 28])

a[:2].shape #第0,1张图
>>>torch.Size([2, 3, 28, 28])

a[:2,:1,:,:].shape #第0,1张图的第1个通道上的数据
>>>torch.Size([2, 1, 28, 28])

a[:2,1:,:,:].shape #第0,1张图的(第一个通道,末尾通道)上的数据
>>>torch.Size([2, 2, 28, 28])

a[:2,-1:,:,:].shape #第0,1张图的末尾通道的数据
>>>torch.Size([2, 1, 28, 28])

a[:,:,0:28:2,0:28:2].shape #隔行采样
torch.Size([4, 3, 14, 14])

a[:,:,::2,::2].shape #同上
>>>torch.Size([4, 3, 14, 14])

2、具体索引

a.shape
>>>torch.Size([4, 3, 28, 28])

a.index_select(0,torch.tensor([0,2])).shape  
#0表示对第一个维度，也就是图片张数进行操作，[0,2]表示第0和第2张图片
>>>torch.Size([2, 3, 28, 28])

a.index_select(1,torch.tensor([1,2])).shape
#1表示对第二个维度，在通道数上，取第1和第2个通道，当然图片数就是全部了
>>>torch.Size([4, 2, 28, 28])

a.index_select(2, torch.arange(14)).shape
>>>torch.Size([4, 3, 14, 28])

a[...].shape #表示所有维度都取
>>>torch.Size([4, 3, 28, 28])

a[0,...,::2].shape 
#第一个维度取0，中间一个维度全取(...表示任意多的维度)，最后一个维度取步长为2 
>>>torch.Size([3, 28, 14])

a[:,1,...].shape
>>>torch.Size([4, 28, 28])

a[...,:2].shape
>>>torch.Size([4, 3, 28, 2])

###select by mask
x = torch.randn(3, 4)
x
>>>tensor([[ 1.1194, -0.5518, -0.2115, -1.4508],
        [ 1.3920,  1.0053, -1.2985, -1.2529],
        [-1.0730,  1.4239,  0.1493, -0.2288]])

mask = x.ge(0.5)  #把元素大于0.5的设为1，小于0.5为0
mask
>>>tensor([[1, 0, 0, 0],
        [1, 1, 0, 0],
        [0, 1, 0, 0]], dtype=torch.uint8)

torch.masked_select(x, mask)
>>>tensor([1.1194, 1.3920, 1.0053, 1.4239])

torch.masked_select(x, mask).shape
>>>torch.Size([4])

#select by flatten index
src = torch.tensor([[4,3,5],
                  [6,7,8]])
torch.take(src, torch.tensor([0,2,5]))
>>>tensor([4, 5, 8])

 三、维度变换

1、view/reshape

#view和reshape
a = torch.rand(4,1,28,28)
a.shape

>>>torch.Size([4, 1, 28, 28])

a.view(4, 28*28) #把后面三个维度合并到一起，每张图片都用一个784维的向量表示
>>>tensor([[0.1820, 0.8625, 0.2687,  ..., 0.2841, 0.4331, 0.6522],
        [0.4611, 0.7675, 0.7003,  ..., 0.9976, 0.9174, 0.2024],
        [0.7160, 0.8296, 0.7346,  ..., 0.6240, 0.6848, 0.3391],
        [0.3076, 0.6013, 0.2066,  ..., 0.2345, 0.5690, 0.2885]])


a.view(4,28*28).shape
>>>torch.Size([4, 784])

a.view(4*28,28).shape #所有照片的所有行都放到一个行向量中
>>>torch.Size([112, 28])

a.view(4*1, 28, 28).shape #所有照片的通道数进行合并
>>>torch.Size([4, 28, 28])

b = a.view(4, 784)
b.view(4, 1, 28, 28).shape
>>>torch.Size([4, 1, 28, 28])

2、unsqueeze/squeeze

a.shape
>>>torch.Size([4, 1, 28, 28])

a.unsqueeze(0).shape #在第0维度前插入一个维度
>>>torch.Size([1, 4, 1, 28, 28])

a.unsqueeze(-1).shape #对应于4
>>>torch.Size([4, 1, 28, 28, 1])

a.unsqueeze(-4).shape #对应于1
>>>torch.Size([4, 1, 1, 28, 28])

a.unsqueeze(4).shape
>>>torch.Size([4, 1, 28, 28, 1])

a.unsqueeze(-5).shape #对应于0
>>>torch.Size([1, 4, 1, 28, 28])

实际数据

a = torch.tensor([1.2,2.3])
a.shape
>>>torch.Size([2])

a.unsqueeze(-1)   #[2] -> [2,1]
>>>tensor([[1.2000],
        [2.3000]])

a.unsqueeze(-1).shape
>>>torch.Size([2, 1])

a.unsqueeze(0)  #[2] -> [1,2]
>>>tensor([[1.2000, 2.3000]])

a.unsqueeze(0).shape
>>>torch.Size([1, 2])

#For example
b = torch.rand(32)
f = torch.rand(4,32,14,14)

b = b.unsqueeze(1).unsqueeze(2).unsqueeze(0)
b.shape
>>>torch.Size([1, 32, 1, 1])

 squeeze

b.shape
>>>torch.Size([1, 32, 1, 1])

b.squeeze().shape  #不给参数，全部解压
>>>torch.Size([32])

b.squeeze(0).shape
>>>torch.Size([32, 1, 1])

b.squeeze(-1).shape
>>>torch.Size([1, 32, 1])

b.squeeze(-1).shape
>>>torch.Size([1, 32, 1])

b.squeeze(1).shape
>>>torch.Size([1, 32, 1, 1])

b.squeeze(-4).shape
>>>torch.Size([32, 1, 1])

expand

a = torch.rand(4,32,14,14)
b.shape

>>>torch.Size([1, 32, 1, 1])

b.expand(4,32,14,14).shape #仅限于1->N
>>>torch.Size([4, 32, 14, 14])

b.expand(-1,32,-1,-1).shape #-1表示该维度保持不变
>>>torch.Size([1, 32, 1, 1])

b.expand(-1,32,-1,4).shape
>>>torch.Size([1, 32, 1, 4])

repeat#维度拷贝

b.shape
>>>torch.Size([1, 32, 1, 1])

b.repeat(4,32,1,1).shape #每一个维度需要拷贝的次数
>>>torch.Size([4, 1024, 1, 1])

b.repeat(4,1,1,1).shape
>>>torch.Size([4, 32, 1, 1])

b.repeat(4,1,32,32).shape
>>>torch.Size([4, 32, 32, 32])

矩阵转置

a = torch.randn(3,4)
a.t() #只能用于二维
>>>tensor([[ 1.4149, -1.2417,  0.4027],
        [-0.3130,  1.0590,  0.1123],
        [-1.3699, -0.3791, -1.0308],
        [-0.8360, -0.2011, -0.4700]])


a = torch.randn(4, 3, 32, 32)

a1 = a.transpose(1,3).contiguous().view(4,3*32*32).view(4,3,32,32)
>>>#[b c H W] ->[b w H C] ->[b C W H]

a2 = a.transpose(1,3).contiguous().view(4,3*32*32).view(4,32,32,3).transpose(1,3)
# [b c H W] -> [b W H c] -> [b WHc] -> [b W H c] -> [b c H W]

a1.shape, a2.shape
>>>(torch.Size([4, 3, 32, 32]), torch.Size([4, 3, 32, 32]))

torch.all(torch.eq(a,a1))
>>>tensor(0, dtype=torch.uint8)

torch.all(torch.eq(a,a2))
>>>tensor(1, dtype=torch.uint8)

a.transpose(1,3).shape
>>>torch.Size([4, 28, 28, 3])

b = torch.rand(4,3,28,32)
b.transpose(1,3).shape
>>>torch.Size([4, 32, 28, 3])

b.transpose(1,3).transpose(1,2).shape
>>>torch.Size([4, 28, 32, 3])

b.permute(0,2,3,1).shape #直接干，[b c H W] -> [b H W c]
>>>torch.Size([4, 28, 32, 3])