Torch和Numpy之间的简单对比

• numpy和torch两者用法十分相似，torch可以说是神经网络中的numpy。

numpy和torch之间的互相转换：

import torch
import numpy as np

np_data = np.arange(6).reshape(2, 3)
print("Numpy 数据：", np_data)
torch_data = torch.tensor([[0, 1, 2], [3, 4, 5]])
print("Torch 数据：", torch_data)

np2torch = torch.from_numpy(np_data)
print("numpy数据转torch数据:", np2torch)
torch2np = torch_data.numpy()
print("torch数据转numpy数据：", torch2np)

• result：

  

numpy和torch的数学运算：

• 详细的API用法请参考官方地址：https://pytorch.org/docs/stable/torch.html

• 中文文档地址：https://pytorch-cn.readthedocs.io/zh/latest/package_references/Tensor/

1. 计算绝对值abs

   import torch
   import numpy as np
   
   data = [-1,2,-3,4]
   np_abs = np.abs(np.array(data))
   print("numpy的abs：", np_abs)
   
   torch_abs = torch.abs(torch.tensor(data))
   print("torch的abs：", torch_abs)
   

   result:

   ​

2. 计算sin torch.tensor和torch.Torch的区别

   import torch
   import numpy as np
   
   data = [-1,2,-3,4]
   np_sin = np.sin(np.array(data))
   print("numpy的sin：", np_sin)
   # 这里改了大写的Tensor
   torch_sin = torch.sin(torch.Tensor(data))
   print("torch的sin：", torch_sin)
   

   result:

   ​

3. 计算均值mean

   import torch
   import numpy as np
   
   data = [-1,2,-3,4]
   np_mean = np.mean(np.array(data))
   print("numpy的均值：", np_mean)
   # 这里改了大写的Tensor
   torch_mean = torch.mean(torch.Tensor(data))
   print("torch的均值：", torch_mean)
   

   result:

   ​

4. 矩阵相乘

   import torch
   import numpy as np
   
   matrix_2x3 = [[1,2,3], [4,5,6]]
   matrix_3x2 = [[1,2],[3,4],[5,6]]
   
   np_matmul = np.matmul(matrix_2x3, matrix_3x2)
   print("numpy的矩阵相乘：", np_matmul)
   torch_matmul = torch.mm(torch.tensor(matrix_2x3), torch.tensor(matrix_3x2))
   print("torch的矩阵相乘：", torch_matmul)
   

   result:

   ​