Pytorch常用创建Tensor方法总结
1、import from numpy / list
方法:torch.from_numpy(ndarray)
常见的初始化有torch.tensor和torch.Tensor
区别:
tensor():通过numpy 或 list 的现有数据初始化
Tensor():1、接收数据的维度(,)shape
2、接收现有的数据[,]
a = np.array([1,2,3]) data = torch.from_numpy(a) print(data) """ 输出: tensor([1, 2, 3], dtype=torch.int32) """ b = np.ones([2,3]) data1 = torch.from_numpy(b) print(data1) """ 输出: tensor([[1., 1., 1.], [1., 1., 1.]], dtype=torch.float64) """ # 参数为 list print(torch.tensor([2.,1.2])) """ 输出: tensor([2.0000, 1.2000]) """
2、未初始化 / 设置默认类型
方法:
torch.empty(size)
torch.FloatTensor(d1,d2,d3)
torch.InrTensor(d1,d2,d3)
torch.set_default_tensor_type(torch.DoubleTensor (设置默认类型)
# 未初始化 data = torch.empty(1) print(data) print(torch.Tensor(2,3)) print(torch.IntTensor(3,4)) """ 输出: tensor([0.]) tensor([[0.0000e+00, 0.0000e+00, 2.8026e-45], [0.0000e+00, 1.4013e-45, 0.0000e+00]]) tensor([[1718379891, 1698963500, 1701013878, 1986356256], [ 744842089, 1633899296, 1416782188, 543518841], [1887007844, 1646275685, 543977327, 1601073006]], dtype=torch.int32) """ print(torch.tensor([1.,2]).type()) torch.set_default_tensor_type(torch.DoubleTensor) print(torch.tensor([1.,2]).type()) """ 输出: torch.FloatTensor torch.DoubleTensor """
3、随机生成
torch.rand(size):产生[0,1]均匀分布的数据
torch.rand_like(input, dtype):接收tensor读取shape再用rand生成
torch.randint(low = 0, high, size):随机生成整数值tensor,范围 [min,max):左闭右开
torch.randn(size):N(0,1)均值为0,方差为1的正态分布(N(u,std))
torch.full(size, fill_value):全部赋予相同的值
torch.normal(means,std,out = None)
返回一个张量,包含从给定参数means,std的离散正态分布中抽取随机数。 均值means是一个张量,包含每个输出元素相关的正态分布的均值。 std是一个张量,包含每个输出元素相关的正态分布的标准差。 均值和标准差的形状不须匹配,但每个张量的元素个数须相同。
参数:
means (Tensor) – 均值
std (Tensor) – 标准差
out (Tensor) – 可选的输出张量
data = torch.rand(3,3) print(data) """ 输出: tensor([[0.0775, 0.2610, 0.0833], [0.7911, 0.6999, 0.6589], [0.4790, 0.6801, 0.6582]]) """ data_like = torch.randn_like(data) print(data_like) """ 输出: tensor([[ 0.6866, 2.5939, -0.2480], [-0.9259, -0.3617, 0.5759], [-1.0179, -1.0938, 0.6426]]) """ print(torch.randint(1,10,[3,3])) """ 输出: tensor([[7, 3, 2], [8, 6, 7], [7, 7, 7]]) """ data = torch.randn(3,3) print(data) """ 输出: tensor([[-0.6225, -0.1253, -0.1083], [-0.3199, -0.5670, 0.2898], [-0.6500, 0.9275, 1.0377]]) """ data = torch.normal(mean=torch.full([10],0),std=torch.arange(1,0,-0.1)) print(data) """ 输出: tensor([-0.6509, -1.4877, 0.4740, 1.1891, 0.1009, -0.4449, -0.3422, 0.1519, -0.2735, 0.1140]) """ print(torch.full([2,4],7)) print(torch.full([],7)) # 标量 """ 输出: tensor([[7., 7., 7., 7.], [7., 7., 7., 7.]]) tensor(7.) """
4、序列生成
torch.arange(start, end, step)
# [start,end) 左闭右开,默认步长为1
torch.range(start, end, step) (已被arange替代)
# 包括end,step是两个点间距
torch.linspace(start, end, steps) # 等差数列
# 包括end, steps 是点的个数,包括端点, (等距离)
torch.logspace(start, end, steps) #
print(torch.arange(0,10)) print(torch.arange(0,10,2)) print(torch.linspace(0,10,steps=3)) print(torch.linspace(0,10,steps=11)) print(torch.logspace(0,-1,steps=10)) print(torch.logspace(0,1,steps=10)) """ 输出: tensor([0, 1, 2, 3, 4, 5, 6, 7, 8, 9]) tensor([0, 2, 4, 6, 8]) tensor([ 0., 5., 10.]) tensor([ 0., 1., 2., 3., 4., 5., 6., 7., 8., 9., 10.]) tensor([1.0000, 0.7743, 0.5995, 0.4642, 0.3594, 0.2783, 0.2154, 0.1668, 0.1292, 0.1000]) tensor([ 1.0000, 1.2915, 1.6681, 2.1544, 2.7826, 3.5938, 4.6416, 5.9948, 7.7426, 10.0000]) """
5、全零、全一、单位矩阵
torch.zeros(size)
torch.zeros_like(input, dtype)
torch.ones(size)
torch.ones_like(input, dtype)
torch.eye(size)
print(torch.ones(3,3)) print(torch.zeros(2,3)) print(torch.eye(4,4)) print(torch.eye(2)) """ 输出: tensor([[1., 1., 1.], [1., 1., 1.], [1., 1., 1.]]) tensor([[0., 0., 0.], [0., 0., 0.]]) tensor([[1., 0., 0., 0.], [0., 1., 0., 0.], [0., 0., 1., 0.], [0., 0., 0., 1.]]) tensor([[1., 0.], [0., 1.]]) """
6、torch.randperm(n) # 生成一个0到n-1的n-1个整数的随机排列
print(torch.randperm(10)) """ 输出: tensor([0, 1, 4, 7, 9, 8, 6, 3, 2, 5]) """