目录
Pytorch使用
1.张量Tensor
张量是一个统称，其中包含很多类型:
1.0阶张量: 标量、常数，0-D Tensor
2.1阶张量:向量，1-D Tensor
3.2阶张量:矩阵，2-D Tensor
4.3阶张量
5. ..
6.N阶张量
2.Pytorch中创建张量
1.使用python中的列表或者序列创建tensor
a = torch.Tensor([[1, 2, 3], [4, 5, 6]])
print(a)
'''
tensor([[1., 2., 3.],
[4., 5., 6.]])
'''2.使用numpy中的数组创建tensor
array = np.arange(12).reshape(3, 4)
print(array)
'''
[[ 0 1 2 3]
[ 4 5 6 7]
[ 8 9 10 11]]
'''3.使用torch的api创建tensor
1.torch.empty(3,4)创建3行4列的空的tensor，会用无用数据进行填充
2.torch.ones([3,4]) 创建3行4列的全为1的tensor
3.torch.zeros([3,4])创建3行4列的全为0的tensor
4.torth.rand([3,4]) 创建3行4列的随机值的tensor，随机值的区间是 [0，1)
rand = torch.rand(2, 3)
print(rand)
'''
tensor([[0.0311, 0.8508, 0.8894],
[0.5359, 0.6662, 0.5128]])
'''5.torch,randint(1ow=0,high=10,size=[3,4])创建3行4列的随机整数的tensor，随机值的区间是[low，high)
randint = torch.randint(3, 10, (2, 2))
print(randint)
'''
tensor([[3, 3],
[5, 6]])
'''6.torch.randn([3,4])创动3行4列的随机数的tensor，随机值的分布式均值为0，方差为1
3. Pytorch中tensor的常用方法
1.获取tensor中的数据(当tensor中只有一个元素可用):tensor .item()
t1 = torch.Tensor([[10]])
print(t1)
print(t1.item())
'''
10
'''2.转化为numpy数组
t2 = t1.numpy()
print(t2)
'''
[[10.]]
'''3.获取形状: tensor.size()
t3 = torch.rand(3, 4)
print(t3.size())
'''
torch.Size([3, 4])
'''4.形状改变: tensor.view((3,4))。类似numpy中的reshape，是一种浅拷贝，仅仅是形状发生改变
t4 = t3.view(2, 6)
print(t3)
print(t4)
'''
tensor([[0.5064, 0.2271, 0.7398, 0.1612],
[0.9324, 0.7501, 0.8088, 0.4055],
[0.7744, 0.1162, 0.8764, 0.2311]])
tensor([[0.5064, 0.2271, 0.7398, 0.1612, 0.9324, 0.7501],
[0.8088, 0.4055, 0.7744, 0.1162, 0.8764, 0.2311]])
'''5.获取维（阶）数: tensor.dim()
print(t4.dim())
'''
2
'''6.获取最大值: tensor.max()
print(t4)
print(t4.max())
'''
tensor([[0.6645, 0.9753, 0.1691, 0.9781, 0.9510, 0.5923],
[0.7698, 0.4558, 0.5393, 0.7076, 0.2258, 0.8842]])
tensor(0.9781)
'''7.转置: tensor.t() （在numpy中为 T）
print(t4)
print(t4.t())
'''
tensor([[0.8161, 0.3213, 0.7988, 0.9154, 0.0466, 0.0426],
[0.3853, 0.7590, 0.3691, 0.5878, 0.1044, 0.6844]])
tensor([[0.8161, 0.3853],
[0.3213, 0.7590],
[0.7988, 0.3691],
[0.9154, 0.5878],
[0.0466, 0.1044],
[0.0426, 0.6844]])
'''t4 = torch.rand(2, 3, 4)
print(t4)
print(t4.transpose(0, 1))
print(t4.permute(1, 0, 2))
'''
tensor([[[0.5269, 0.0171, 0.0110, 0.3139],
[0.9625, 0.0424, 0.1559, 0.9512],
[0.2839, 0.9033, 0.7233, 0.7983]],
[[0.4119, 0.1300, 0.7744, 0.7593],
[0.7209, 0.8706, 0.0809, 0.3257],
[0.7994, 0.2252, 0.3721, 0.4984]]])
tensor([[[0.5269, 0.0171, 0.0110, 0.3139],
[0.4119, 0.1300, 0.7744, 0.7593]],
[[0.9625, 0.0424, 0.1559, 0.9512],
[0.7209, 0.8706, 0.0809, 0.3257]],
[[0.2839, 0.9033, 0.7233, 0.7983],
[0.7994, 0.2252, 0.3721, 0.4984]]])
tensor([[[0.5269, 0.0171, 0.0110, 0.3139],
[0.4119, 0.1300, 0.7744, 0.7593]],
[[0.9625, 0.0424, 0.1559, 0.9512],
[0.7209, 0.8706, 0.0809, 0.3257]],
[[0.2839, 0.9033, 0.7233, 0.7983],
[0.7994, 0.2252, 0.3721, 0.4984]]])
'''8.tensor[1,3] 获取tensor中第一行第三列的值
t4 = torch.rand(2, 3, 2)
print(t4)
print(t4[1, 2])
'''
tensor([[[0.0768, 0.1384],
[0.9043, 0.0226],
[0.0334, 0.2946]],
[[0.9075, 0.4634],
[0.8371, 0.1950],
[0.3811, 0.9326]]])
tensor([0.3811, 0.9326])
'''9.tensor[1,2]=100 对tensor中第一行第三列的位置进行赋值100
t4 = torch.rand(2, 3, 2)
print(t4)
print(t4[1, 2])
t4[1, 2] = 100
print(t4)
'''
tensor([[[0.2413, 0.9267],
[0.7468, 0.1347],
[0.8241, 0.6128]],
[[0.1088, 0.0445],
[0.5081, 0.5531],
[0.4178, 0.0222]]])
tensor([0.4178, 0.0222])
tensor([[[2.4131e-01, 9.2668e-01],
[7.4677e-01, 1.3468e-01],
[8.2412e-01, 6.1279e-01]],
[[1.0878e-01, 4.4516e-02],
[5.0811e-01, 5.5308e-01],
[1.0000e+02, 1.0000e+02]]])
'''10.tensor的切片
4. tensor的数据类型
tensor中的数据类型非常多，常见类型如下：
上图中的Tensor types 表示这种type的tensor是实例
1. 获取tensor的数据类型：tensor.dtype
print(t4)
print(t4.dtype)
'''
tensor([[[ 0.9967, 0.8391],
[ 0.9651, 0.7411],
[ 0.4698, 0.5371]],
[[ 0.9868, 0.1172],
[ 0.3568, 0.8628],
[100.0000, 100.0000]]])
torch.float32
'''2. 创建数据的时候指定类型
t5 = torch.ones([2, 3], dtype=torch.float32)
print(t5)
print(t5.dtype)
'''
tensor([[1., 1., 1.],
[1., 1., 1.]])
torch.float32
'''3. 类型的修改
t5 = torch.ones([2, 3], dtype=torch.int32)
print(t5.dtype)
t5 = t5.type(torch.float)
print(t5.dtype)
t5 = t5.double()
print(t5.dtype)
'''
torch.int32
torch.float32
torch.float64
'''5. tensor的其他操作
1. tensor和tensor相加
x = torch.rand(3, 4)
x = x.new_ones((2, 6), dtype=torch.float32)
y = torch.rand(2, 6)
print(x)
print(y)
print(x + y)
'''
tensor([[1., 1., 1., 1., 1., 1.],
[1., 1., 1., 1., 1., 1.]])
tensor([[0.0763, 0.2841, 0.7078, 0.7183, 0.9554, 0.5018],
[0.0874, 0.9510, 0.0774, 0.1421, 0.1593, 0.8375]])
tensor([[1.0763, 1.2841, 1.7078, 1.7183, 1.9554, 1.5018],
[1.0874, 1.9510, 1.0774, 1.1421, 1.1593, 1.8375]])
'''
print(torch.add(x, y))
'''
tensor([[1.0763, 1.2841, 1.7078, 1.7183, 1.9554, 1.5018],
[1.0874, 1.9510, 1.0774, 1.1421, 1.1593, 1.8375]])
'''
print(x.add(y))
'''
tensor([[1.0763, 1.2841, 1.7078, 1.7183, 1.9554, 1.5018],
[1.0874, 1.9510, 1.0774, 1.1421, 1.1593, 1.8375]])
'''
2. 原地修改的方法
print(x.add_(y)) # 带下划线的方法会对x进行就地修改
'''
tensor([[1.0763, 1.2841, 1.7078, 1.7183, 1.9554, 1.5018],
[1.0874, 1.9510, 1.0774, 1.1421, 1.1593, 1.8375]])
'''
print(x)
'''
tensor([[1.0763, 1.2841, 1.7078, 1.7183, 1.9554, 1.5018],
[1.0874, 1.9510, 1.0774, 1.1421, 1.1593, 1.8375]])
'''
3. Gpu中的tensor的使用
- 实例化device: torch.device("cuda:0"iftorch.cuda.is avaiable() else "cpu")
- tensor.to(device) #把tensor转化为CUDA支持的tensor，或者cpu支持的tensor