Welcome to OGeek Q&A Community for programmer and developer-Open, Learning and Share
Welcome To Ask or Share your Answers For Others

Categories

0 votes
394 views
in Technique[技术] by (71.8m points)

python 3.x - np.add.at indexing with array

I'm working on cs231n and I'm having a difficult time understanding how this indexing works. Given that

x = [[0,4,1], [3,2,4]]
dW = np.zeros(5,6)
dout = [[[  1.19034710e-01  -4.65005990e-01   8.93743168e-01  -9.78047129e-01
            -8.88672957e-01  -4.66605091e-01]
         [ -1.38617461e-03  -2.64569728e-01  -3.83712733e-01  -2.61360826e-01
            8.07072009e-01  -5.47607277e-01]
         [ -3.97087458e-01  -4.25187949e-02   2.57931759e-01   7.49565950e-01
           1.37707667e+00   1.77392240e+00]]

       [[ -1.20692745e+00  -8.28111550e-01   6.53041092e-01  -2.31247762e+00
         -1.72370321e+00   2.44308033e+00]
        [ -1.45191870e+00  -3.49328154e-01   6.15445782e-01  -2.84190582e-01
           4.85997687e-02   4.81590106e-01]
        [ -1.14828583e+00  -9.69055406e-01  -1.00773809e+00   3.63553835e-01
          -1.28078363e+00  -2.54448436e+00]]]

The operation they do is

np.add.at(dW, x, dout)

x is a two dimensional array. How does indexing work here? I went through np.ufunc.at documentation but they have simple examples with 1d array and constant:

np.add.at(a, [0, 1, 2, 2], 1)
See Question&Answers more detail:os

与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…
Welcome To Ask or Share your Answers For Others

1 Reply

0 votes
by (71.8m points)
In [226]: x = [[0,4,1], [3,2,4]]
     ...: dW = np.zeros((5,6),int)

In [227]: np.add.at(dW,x,1)
In [228]: dW
Out[228]: 
array([[0, 0, 0, 1, 0, 0],
       [0, 0, 0, 0, 1, 0],
       [0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0],
       [0, 0, 1, 0, 0, 0]])

With this x there aren't any duplicate entries, so add.at is the same as using += indexing. Equivalently we can read the changed values with:

In [229]: dW[x[0], x[1]]
Out[229]: array([1, 1, 1])

The indices work the same either way, including broadcasting:

In [234]: dW[...]=0
In [235]: np.add.at(dW,[[[1],[2]],[2,4,4]],1)
In [236]: dW
Out[236]: 
array([[0, 0, 0, 0, 0, 0],
       [0, 0, 1, 0, 2, 0],
       [0, 0, 1, 0, 2, 0],
       [0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0]])

possible values

The values have to be broadcastable, with respect to the indexes:

In [112]: np.add.at(dW,[[[1],[2]],[2,4,4]],np.ones((2,3)))
...
In [114]: np.add.at(dW,[[[1],[2]],[2,4,4]],np.ones((2,3)).ravel())
...
ValueError: array is not broadcastable to correct shape
In [115]: np.add.at(dW,[[[1],[2]],[2,4,4]],[1,2,3])

In [117]: np.add.at(dW,[[[1],[2]],[2,4,4]],[[1],[2]])

In [118]: dW
Out[118]: 
array([[ 0,  0,  0,  0,  0,  0],
       [ 0,  0,  3,  0,  9,  0],
       [ 0,  0,  4,  0, 11,  0],
       [ 0,  0,  0,  0,  0,  0],
       [ 0,  0,  0,  0,  0,  0]])

In this case the indices define a (2,3) shape, so (2,3),(3,), (2,1), and scalar values work. (6,) does not.

In this case, add.at is mapping a (2,3) array onto a (2,2) subarray of dW.


与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…
OGeek|极客中国-欢迎来到极客的世界,一个免费开放的程序员编程交流平台!开放,进步,分享!让技术改变生活,让极客改变未来! Welcome to OGeek Q&A Community for programmer and developer-Open, Learning and Share
Click Here to Ask a Question

...