- #Tensor索引操作
- '''''
- Tensor支持与numpy.ndarray类似的索引操作,语法上也类似
- 如无特殊说明,索引出来的结果与原tensor共享内存,即修改一个,另一个会跟着修改
- '''
- import torch as t
- a = t.randn(3,4)
- '''''tensor([[ 0.1986, 0.1809, 1.4662, 0.6693],
- [-0.8837, -0.0196, -1.0380, 0.2927],
- [-1.1032, -0.2637, -1.4972, 1.8135]])'''
- print(a[0]) #第0行
- '''''tensor([0.1986, 0.1809, 1.4662, 0.6693])'''
- print(a[:,0]) #第0列
- '''''tensor([ 0.1986, -0.8837, -1.1032])'''
- print(a[0][2]) #第0行第2个元素,等价于a[0,2]
- '''''tensor(1.4662)'''
- print(a[0][-1]) #第0行最后一个元素
- '''''tensor(0.6693)'''
- print(a[:2,0:2]) #前两行,第0,1列
- '''''tensor([[ 0.1986, 0.1809],
- [-0.8837, -0.0196]])'''
- print(a[0:1,:2]) #第0行,前两列
- '''''tensor([[0.1986, 0.1809]])'''
- print(a[0,:2]) #注意两者的区别,形状不同
- '''''tensor([0.1986, 0.1809])'''
- print(a>1)
- '''''tensor([[0, 0, 1, 0],
- [0, 0, 0, 0],
- [0, 0, 0, 1]], dtype=torch.uint8)'''
- print(a[a>1]) #等价于a.masked_select(a>1),选择结果与原tensor不共享内存空间
- print(a.masked_select(a>1))
- '''''tensor([1.4662, 1.8135])
- tensor([1.4662, 1.8135])'''
- print(a[t.LongTensor([0,1])])
- '''''tensor([[ 0.1986, 0.1809, 1.4662, 0.6693],
- [-0.8837, -0.0196, -1.0380, 0.2927]])'''
- '''''
- 常用的选择函数
- index_select(input,dim,index) 在指定维度dim上选取,列如选择某些列、某些行
- masked_select(input,mask) 例子如上,a[a>0],使用ByteTensor进行选取
- non_zero(input) 非0元素的下标
- gather(input,dim,index) 根据index,在dim维度上选取数据,输出size与index一样
- gather是一个比较复杂的操作,对一个二维tensor,输出的每个元素如下:
- out[i][j] = input[index[i][j]][j] #dim = 0
- out[i][j] = input[i][index[i][j]] #dim = 1
- '''
- b = t.arange(0,16).view(4,4)
- '''''tensor([[ 0, 1, 2, 3],
- [ 4, 5, 6, 7],
- [ 8, 9, 10, 11],
- [12, 13, 14, 15]])'''
- index = t.LongTensor([[0,1,2,3]])
- print(b.gather(0,index)) #取对角线元素
- '''''tensor([[ 0, 5, 10, 15]])'''
- index = t.LongTensor([[3,2,1,0]]).t() #取反对角线上的元素
- print(b.gather(1,index))
- '''''tensor([[ 3],
- [ 6],
- [ 9],
- [12]])'''
- index = t.LongTensor([[3,2,1,0]]) #取反对角线的元素,与上面不同
- print(b.gather(0,index))
- '''''tensor([[12, 9, 6, 3]])'''
- index = t.LongTensor([[0,1,2,3],[3,2,1,0]]).t()
- print(b.gather(1,index))
- '''''tensor([[ 0, 3],
- [ 5, 6],
- [10, 9],
- [15, 12]])'''
- '''''
- 与gather相对应的逆操作是scatter_,gather把数据从input中按index取出,而
- scatter_是把取出的数据再放回去,scatter_函数时inplace操作
- out = input.gather(dim,index)
- out = Tensor()
- out.scatter_(dim,index)
- '''
- x = t.rand(2, 5)
- print(x)
- c = t.zeros(3, 5).scatter_(0, t.LongTensor([[0, 1, 2, 0, 0], [2, 0, 0, 1, 2]]), x)
- print(c)
- 2018-10-23 20:30:30