用如下 a b 进行运算演示

>>> a                    
tensor([[0.7967, 0.5056, 0.7963],[0.8603, 0.7029, 0.7590]])
>>> b
tensor([[0.6923, 0.0411, 0.8713],[0.0483, 0.2948, 0.3286]])

• 一般加减乘除运算： add/mimus/multiply/divide

  >>> a+b  ## torch.add(a,b)
  tensor([[1.4890, 0.5467, 1.6676],[0.9086, 0.9977, 1.0876]])
  >>> a-b  ## torch.sub(a,b)
  tensor([[ 0.1044,  0.4646, -0.0750],[ 0.8120,  0.4080,  0.4304]])
  >>> a*b  ## torch.mul(a,b)
  tensor([[0.5516, 0.0208, 0.6938],[0.0416, 0.2072, 0.2494]])
  >>> a/b  ## torch.div(a,b)
  tensor([[ 1.1508, 12.3163,  0.9139],[17.8087,  2.3840,  2.3099]])
  

• 矩阵相乘: matmul
  • 对于大于二维的，只取后面的二维进行运算

  >>> a@b.t()  ## torch.matmul(a, b.t()) 
  tensor([[1.2662, 0.4492],[1.2858, 0.4982]])
  

• pow，sqrt, exp, log

  >>> a**2  ## a.pow(2)
  tensor([[0.6348, 0.2557, 0.6341],[0.7401, 0.4940, 0.5761]])
  >>> a**0.5  ## a.sqrt()
  tensor([[0.8926, 0.7111, 0.8924],[0.9275, 0.8384, 0.8712]])
  >>> torch.exp(a)  ## e^a
  tensor([[2.2183, 1.6580, 2.2173],[2.3638, 2.0196, 2.1362]])
  >>> torch.log(a)  ## log_e a 
  tensor([[-0.2272, -0.6819, -0.2278],[-0.1505, -0.3526, -0.2757]])
  

• 近似计算：floor, ceil, round 和 裁剪 clamp

  >>> a.floor()  ## 向上取整
  tensor([[0., 0., 0.],[0., 0., 0.]])
  >>> a.ceil()  ## 向下取整
  tensor([[1., 1., 1.],[1., 1., 1.]])
  >>> a.round()  ## 四舍五入
  tensor([[1., 1., 1.],[1., 1., 1.]])>>> a.clamp(0.1, 0.8)  ## 裁剪到指定区间 0.1 < a < 0.8
  tensor([[0.7967, 0.5056, 0.7963],[0.8000, 0.7029, 0.7590]])>>> a.trunc()  ## 拆分获得整数部分
  tensor([[0., 0., 0.],[0., 0., 0.]])
  >>> a.frac() 	 ## 拆分获得小数部分
  tensor([[0.7967, 0.5056, 0.7963],[0.8603, 0.7029, 0.7590]])
  

---

• B站视频参考资料