numpy-与array.sum相反（减法？）

Question

I was wondering if there a method that can subtract values? Something similar to sum

for Examples

> np.array([[10, 2], [1, 2]]).sum() 
15
" imaginary method "
> np.array([[10, 2], [1, 2]]).sub() 
6

# axis = 1
> np.array([[10, 2], [1, 2]]).sum(axis=1)
array([12,  3])
" imaginary method "
> np.array([[10, 2], [1, 2]]).sum(axis=1)
array([8,  -1])

# axis = 0
> np.array([[10, 2], [1, 2]]).sum(axis=0)
array([11,  4])
"imaginary"
> np.array([[10, 2], [1, 2]]).sub(axis=0)
array([9,  0])

I am frustrated I cann't find anything in docs (somehow numpy docs are not easy to use if you don't know what you looking for).

thank you.

Answer 1

The difference equivalent of np.sum() is np.diff()

Docs

Answer 2

np.sum is np.add.reduce :

In [87]: np.add.reduce(arr, axis=0)                                             
Out[87]: array([11,  4])
In [88]: np.add.reduce(arr, axis=1)                                             
Out[88]: array([12,  3])

There is a subtract ufunc too:

In [93]: np.subtract.reduce(arr, axis=0)                                        
Out[93]: array([9, 0])
In [94]: np.subtract.reduce(arr, axis=1)                                        
Out[94]: array([ 8, -1])

np.diff does sliced subtraction:

In [97]: np.subtract(arr[:-1,:], arr[1:,:])                                     
Out[97]: array([[9, 0]])
In [98]: np.subtract(arr[:,:-1], arr[:,1:])                                     
Out[98]: 
array([[ 8],
       [-1]])

For two elements diff and subtract.reduce do the same thing. What's supposed to happen when you have more than 2 rows or columns?

In [109]: arr = np.array([[10, 2, 3], [1, 2, 4], [0, 1, 2]])                    
In [110]: arr                                                                   
Out[110]: 
array([[10,  2,  3],
       [ 1,  2,  4],
       [ 0,  1,  2]])

diff does pair wise subtraction, row 0 from 1, row 1 from 2:

In [111]: np.diff(arr, axis=0)                                                  
Out[111]: 
array([[-9,  0,  1],
       [-1, -1, -2]])

subtract.reduce does a cumulative, which might be easier to follow with the accumulate alternative:

In [112]: np.subtract.reduce(arr, axis=0)                                       
Out[112]: array([ 9, -1, -3])
In [113]: np.subtract.accumulate(arr, axis=0)                                   
Out[113]: 
array([[10,  2,  3],
       [ 9,  0, -1],
       [ 9, -1, -3]])