『PyTorch』第五弹_深入理解Tensor对象_上：初始化以及尺寸调整
一、创建Tensor
特殊方法：
t.arange(1,6,2)
t.linspace(1,10,3)
t.randn(2,3) # 标准分布，*size t.randperm(5) # 随机排序，从0到n t.normal(means=t.arange(0, 11), std=t.arange(1, 0, -0.1))
概览：
"""创建空Tensor""" a = t.Tensor(2, 3) # 创建和b大小一致的Tensor c = t.Tensor(a.size()) print(a,c) # 数值取决于内存空间状态
-9.6609e+30 7.9594e-43 -4.1334e+27 7.9594e-43 -4.1170e+27 7.9594e-43 [torch.FloatTensor of size 2x3] -9.6412e+30 7.9594e-43 -9.6150e+30 7.9594e-43 -4.1170e+27 7.9594e-43 [torch.FloatTensor of size 2x3]
"""由list/tuple创建Tensor""" b = t.Tensor([[1,2,3],[4,5,6]]) print(b) # 根据list初始化Tensor print(b.tolist()) print(b) # 并非inplace转换
1 2 3 4 5 6 [torch.FloatTensor of size 2x3] [[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]] 1 2 3 4 5 6 [torch.FloatTensor of size 2x3]
# 等价写法，查看元素个数(2*3=6) print(b.numel()) print(b.nelement())
6 6
# 传入tuple等价于传入list d = t.Tensor((2,3)) print(d)
2 3 [torch.FloatTensor of size 2]
"""创建特定Tensor""" print(t.eye(2,3)) print(t.ones(2,3)) print(t.zeros(2,3)) print(t.arange(1,6,2)) print(t.linspace(1,10,3)) # 几个特殊初始化方法 print(t.randn(2,3)) # 标准分布，*size print(t.randperm(5)) # 随机排序，从0到n print(t.normal(means=t.arange(0, 11), std=t.arange(1, 0, -0.1)))
1 0 0 0 1 0 [torch.FloatTensor of size 2x3] 1 1 1 1 1 1 [torch.FloatTensor of size 2x3] 0 0 0 0 0 0 [torch.FloatTensor of size 2x3] 1 3 5 [torch.FloatTensor of size 3] 1.0000 5.5000 10.0000 [torch.FloatTensor of size 3] -0.9959 -0.8446 0.7241 3.0315 -0.5367 1.0722 [torch.FloatTensor of size 2x3] 4 3 2 1 0 [torch.LongTensor of size 5] -0.5880 1.2708 1.5530 3.2490 4.7693 4.9497 6.0663 6.1482 7.9109 8.9492 10.0000 [torch.FloatTensor of size 11]
二、尺度调整
特殊方法：
a.view(-1,3） b.unsqueeze_(0) b.resize_(3,3)
概览：
a = t.arange(0,6) print(a.view(2,3)) # 非inplace print(a.view(-1,3)) # -1为自动计算大小
0 1 2 3 4 5 [torch.FloatTensor of size 2x3] 0 1 2 3 4 5 [torch.FloatTensor of size 2x3]
b = a.view(-1,3) b.unsqueeze_(0) print(b) print(b.size())
(0 ,.,.) = 0 1 2 3 4 5 [torch.FloatTensor of size 1x2x3] torch.Size([1, 2, 3])
c = b.view(1,1,1,2,3) print(c.squeeze_(0)) # 压缩第0个1 print(c.squeeze_()) # 压缩全部的1
(0 ,0 ,.,.) = 0 1 2 3 4 5 [torch.FloatTensor of size 1x1x2x3] 0 1 2 3 4 5 [torch.FloatTensor of size 2x3]
# view要求前后元素数相同，resize_没有这个要求 # resize_没有对应的非inplace操作版本 print(b.resize_(1,3)) print(b.resize_(3,3)) print(b)
0 1 2 [torch.FloatTensor of size 1x3] 0.0000e+00 1.0000e+00 2.0000e+00 3.0000e+00 4.0000e+00 5.0000e+00 3.3845e+15 0.0000e+00 0.0000e+00 [torch.FloatTensor of size 3x3] 0.0000e+00 1.0000e+00 2.0000e+00 3.0000e+00 4.0000e+00 5.0000e+00 3.3845e+15 0.0000e+00 0.0000e+00 [torch.FloatTensor of size 3x3]
