I want to match the result of weighted linear sum of vectors with Numpy and Chainer

To sum vectors by weighing them with scalars Use the variable number and chain to
I wrote the following code.

import numpy as np
from chain import Variable
import chain.functions as F

a=np.array([10],[100],[1000]], dtype=np.float32)# set of weights
x = np.array([1, 2, 3], [4, 5, 6], [7, 8, 9]], dtype = np.float32)# collection of vectors
print sum(a*x)# weighted sum of vectors

a = Variable(a)
x = Variable(x)
print F.sum(a*x)

If it's Numpy, it's calculated correctly. If you convert it to Variable, you will get in trouble if it doesn't fit.

What code should I write?
Can you return the same result for numpy and chain?

Thank you for your cooperation.

python chainer

2022-09-30 21:18

1 Answers

Use chain.functions.batch_matmul.

>>print(a*x)
[[   10.    20.    30.]
 [  400.   500.   600.]
 [ 7000.  8000.  9000.]]
>>print(sum(a*x))
[ 7410.  8520.  9630.]
>> print (F.batch_matmul(a,x,transb=True).data)
[[[   10.    20.    30.]]

 [[  400.   500.   600.]]

 [[ 7000.  8000.  9000.]]]
>>print(sum(F.batch_matmul(a,x,transb=True).data))
[[ 7410.  8520.  9630.]]
>>print(F.sum(F.batch_matmul(a,x,transb=True).data),axis=0).data)#sum is also required on the chain side
[[ 7410.  8520.  9630.]]

You can see that the a*x and sum vectors have been calculated.

2022-09-30 21:18

If you have any answers or tips

Popular Tags

python x 4647

android x 1593

java x 1494

javascript x 1427

c x 927

c++ x 878

ruby-on-rails x 696

php x 692

python3 x 685

html x 656