深度学习--PyTorch定义Tensor
一、创建Tensor
1.1未初始化的方法
- torch.empty(：返回一个没有初始化的Tensor，默认是FloatTensor类型。
#torch.empty(d1,d2,d3函数输入的是shape
torch.empty(2,3,5
#tensor([[[-1.9036e-22, 6.8944e-43, 0.0000e+00, 0.0000e+00, -1.0922e-20],
# [ 6.8944e-43, -2.8812e-24, 6.8944e-43, -5.9272e-21, 6.8944e-43],
# [ 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00]],
#
# [[ 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00],
# [ 0.0000e+00, 0.0000e+00, 1.4013e-45, 0.0000e+00, 0.0000e+00],
# [ 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00, 0.0000e+00]]]
- torch.FloatTensor(：返回没有初始化的FloatTensor。
#torch.FloatTensor(d1,d2,d3
torch.FloatTensor(2,2
#tensor([[-0.0000e+00, 4.5907e-41],
# [-7.3327e-21, 6.8944e-43]]
- torch.IntTensor(：返回没有初始化的IntTensor。
#torch.IntTensor(d1,d2,d3
torch.IntTensor(2,2
#tensor([[ 0, 1002524760],
# [-1687359808, 492]], dtype=torch.int32
1.2 随机初始化
随机均匀分布：rand/rand_like,randint
#torch.rand(d1,d2,d3
torch.rand(2,2
#tensor([[0.8670, 0.6158],
# [0.0895, 0.2391]]
#rand_like(
a=torch.rand(3,2
torch.rand_like(a
#tensor([[0.2846, 0.3605],
# [0.3359, 0.2789],
# [0.5637, 0.6276]]
#randint(min,max,[d1,d2,d3]
torch.randint(1,10,[3,3,3]
#tensor([[[3, 3, 8],
# [2, 7, 7],
# [6, 5, 9]],
#
# [[7, 9, 9],
# [6, 3, 9],
# [1, 5, 6]],
#
# [[5, 4, 8],
# [7, 1, 2],
# [3, 4, 4]]]
随机正态分布 randn
randn返回一组符合N(0,1正态分布的随机数据
#randn(d1,d2,d3
torch.randn(2,2
#tensor([[ 0.3729, 0.0548],
# [-1.9443, 1.2485]]
#normal(mean,std 需要给出均值和方差
torch.normal(mean=torch.full([10],0.,std=torch.arange(1,0,-0.1
#tensor([-0.8547, 0.1985, 0.1879, 0.7315, -0.3785, -0.3445, 0.7092, 0.0525, 0.2669, 0.0744]
#后面需要用reshape修正成自己想要的形状
1.3 赋值初始化
- full：返回一个定值
#full([d1,d2,d3],num
torch.full([2,2],6
#tensor([[6, 6],
# [6, 6]]
torch.full([],6
#tensor(6 标量
torch.full([1],6
#tensor([6] 向量
- arange：返回一组阶梯,等差数列
#torch.arange(min,max,step:返回一个[min,max,步长为step的集体数组，默认为1
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]
- linspace/logspace:返回一组阶梯
#torch.linspace(min,max,steps:返回一个[min,max],数量为steps的数组
torch.linspace(1,10,11
#tensor([ 1.0000, 1.9000, 2.8000, 3.7000, 4.6000, 5.5000, 6.4000, 7.3000,
# 8.2000, 9.1000, 10.0000]
#torch.logspace(a,b,steps:返回一个[10^a,10^b]，数量为steps的数组
torch.logspace(0,1,10
#tensor([ 1.0000, 1.2915, 1.6681, 2.1544, 2.7826, 3.5938, 4.6416, 5.9948,
# 7.7426, 10.0000]
- ones/zeros/eye:返回全1全0或者对角阵 ones_like/zeros_like
#torch.ones(d1,d2
torch.ones(2,2
#tensor([[1., 1.],
# [1., 1.]]
#torch.zeros(d1,d2
torch.zeros(2,2
#tensor([[0., 0.],
# [0., 0.]]
#torch.eye( 只能接收一个或两个参数
torch.eye(3
#tensor([[1., 0., 0.],
# [0., 1., 0.],
# [0., 0., 1.]]
torch.eye(2,3
#tensor([[1., 0., 0.],
# [0., 1., 0.]]
1.4 随机打散变量
- randperm：一般用于位置操作。类似random.shuffle(。
torch.randperm(8
#tensor([2, 6, 7, 5, 3, 4, 1, 0]
二、索引与切片
- 简单索引方式
a=torch.rand(4,3,28,28
a[0].shape
#torch.Size([3, 28, 28]
a[0,0,0,0]
#tensor(0.9373
- 批量索引方式 开始位置：结束位置 左边取的到，右边取不到 算是一种切片 [0,1,2]->[-3,-2,-1]
a[:2].shape
#torch.Size([2, 3, 28, 28]
a[1:].shape
#torch.Size([3, 3, 28, 28]
- 隔行采样方式 开始位置:结束位置:间隔
a[:,:,0:28:2,:].shape
#torch.Size([4, 3, 14, 28]
- 任意取样方式 a.index_select(d,[d层的数据索引]
a.index_select(0,torch.tensor([0,2].shape
#torch.Size([2, 3, 28, 28]
a.index_select(1,torch.tensor([0,2].shape
#torch.Size([4, 2, 28, 28]
- ...任意维度取样
a[...].shape
#torch.Size([4, 3, 28, 28]
a[0,...].shape
#torch.Size([3, 28, 28]
a[:,2,...].shape
#torch.Size([4, 28, 28]
- 掩码索引mask x.ge(0.5 表示大于等于0.5的为1，小于0.5的为0
#torch.masked_select 取出掩码对应位置的值
x=torch.randn(3,4
mask=x.ge(0.5
torch.masked_select(x,mask
#tensor([1.6950, 1.2207, 0.6035]
- 具体索引 take(变量，位置 会把变量变为一维的
x=torch.randn(3,4
torch.take(x,torch.tensor([0,1,5]
#tensor([-2.2092, -0.2652, 0.4848]