[Python] Take the index of the largest values in the Numpy multidimensional array

• First look at the two-dimensional array

  a = np.array([[1, 7, 5],
                [8, 2, 13]])
  

  Suppose we want to get the position index of the largest two values
  • We first expand the two-dimensional array into a one-dimensional array, and obtain the sorted index subscript, that is

    index = np.argsort(a.ravel())[:-3:-1]
    
    得到 index 值为 [5 3]
    

  • Next, the index obtained in one dimension is mapped to the high dimension, and the position index in the high dimension array is obtained

    pos = np.unravel_index(index, a.shape)
    
    得到 pos 值为 (array([1, 1], dtype=int64), array([2, 0], dtype=int64))
    

  • Merge pos by column

    np.column_stack(pos)
    
    结果：[[1 2]
    		[1 0]]
    

  • The index [1 2]corresponds to the maximum value 13, and [1 0]the corresponding value is the second largest value8
• Try three-dimensional, three-dimensional (multi-dimensional) and two-dimensional steps are the same

  a = np.array([
          [[1, 7, 5],
           [8, 2, 13]],
          [[25, 0, 3],
           [50, 14, 28]]
      ])
  

  index = np.argsort(a.ravel())[:-3:-1]
  pos = np.unravel_index(index, a.shape)
  print(a.ravel())
  print(index)
  print(pos)
  print(np.column_stack(pos))
  

  a.ravel(): [ 1  7  5  8  2 13 25  0  3 50 14 28]
  
  index: [ 9 11]
  
  pos: (array([1, 1], dtype=int64), array([1, 1], dtype=int64), array([0, 2], dtype=int64))
  
  np.column_stack(pos): [[1 1 0]
   						[1 1 2]]