pytorch安裝與入門（二）--tensor的計算操作

Math operation 數學運算

1. Add/minus/multiply/divide
2. Matmul
3. Pow
4. Sqrt/rsqrt
5. Round

加減乘除

>>> a=torch.rand(3,4)
>>> b=torch.rand(3)
>>> a+b
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
RuntimeError: The size of tensor a (4) must match the size of tensor b (3) at non-singleton dimension 1
>>>
>>> b=torch.rand(4)
>>> a+b
tensor([[1.0229, 1.0478, 1.5048, 0.1652],
        [0.7158, 1.5854, 0.9578, 1.0546],
        [0.8643, 1.6602, 1.1447, 0.4042]])
>>> a-b
tensor([[ 0.8146, -0.9272, -0.3113, -0.1604],
        [ 0.5075, -0.3895, -0.8584,  0.7290],
        [ 0.6560, -0.3147, -0.6715,  0.0786]])
>>> torch.all(torch.eq(a-b,torch.sub(a,b)))
tensor(True)
>>>
>>> torch.all(torch.eq(a+b,torch.add(a,b)))
tensor(True)
>>> torch.all(torch.eq(a*b,torch.mul(a,b)))
tensor(True)
>>> torch.all(torch.eq(a/b,torch.div(a,b)))
tensor(True)

matmul矩陣乘積

 只能對 2d張量（矩陣）進行使用。

>>> a
tensor([[0.5370, 0.6411],
        [0.6101, 0.4873]])
>>> b
tensor([[0.6591, 0.9531],
        [0.9422, 0.0774]])
>>> a*b
tensor([[0.3539, 0.6110],
        [0.5749, 0.0377]])
>>> a@b
tensor([[0.9580, 0.5614],
        [0.8612, 0.6192]])
>>> torch.all(torch.eq(torch.matmul(a,b),a@b))
tensor(True)

pow指數

>>> torch.full([2,2],3)
tensor([[3., 3.],
        [3., 3.]])
>>> torch.full([2,2],3)**3
tensor([[27., 27.],
        [27., 27.]])
>>> torch.full([2,2],3).pow(2)
tensor([[9., 9.],
        [9., 9.]])
>>> torch.full([2,2],3).pow(2).sqrt()
tensor([[3., 3.],
        [3., 3.]])
>>> torch.full([2,2],3).pow(2).rsqrt()
tensor([[0.3333, 0.3333],
        [0.3333, 0.3333]])

note:rsqrt倒數平方根

exp log

>>> torch.exp(torch.full([2,2],1))
tensor([[2.7183, 2.7183],
        [2.7183, 2.7183]])

>>> torch.log((torch.exp(torch.full([2,2],1))))
tensor([[1., 1.],
        [1., 1.]])

Approximation

• floor() .ceil()
• round()
• trunc() .frac()

>>> a=torch.tensor(3.14)
>>> a.floor(),a.ceil(),a.trunc(),a.round(),a.frac()
(tensor(3.), tensor(4.), tensor(3.), tensor(3.), tensor(0.1400))

clamp
gradient clipping

>>> a=torch.rand(2,2)*14
>>> a
tensor([[ 1.3472,  5.9060],
        [12.0558,  4.2571]])
>>> a.clamp(10)
tensor([[10.0000, 10.0000],
        [12.0558, 10.0000]])
>>> a.clamp(0,10)
tensor([[ 1.3472,  5.9060],
        [10.0000,  4.2571]])

 

statistics 統計屬性

▪ norm
▪ mean sum
▪ prod
▪ max, min, argmin, argmax
▪ kthvalue, topk