Python - cross array multiplication

Question

I'm developing a model based on neural network principals. I have an entry layer, weights and an output layer:

[1,2] -- [ [1,1] , [1,1] ] --> [3,3]

My question is whether Python has a simple way (with numpy) to compute the output layers without doing loops and loops.

The current implementation is:

for i in range(0,number_of_out_neurons):
    out_neuron_adder_toWrap = weights[i] * all_input_layer
    out_neuron[i] = sum(out_neuron_adder)         <-- wrapping

Could you implement the loopy version, so that we would have a better idea on what exactly you have in mind and would additionally helps us cross-check against any vectorized approach that could be suggested? — Divakar
– Divakar, Commented Jun 6, 2016 at 17:40
Considering this is just an addition of each input multiplied by each weight, you can just do a dot-product. But this will only work if the input dimensions work with the weight dimensions — Chrispresso
– Chrispresso, Commented Jun 6, 2016 at 17:50
Could you also add their shape information (shapes of input arrays)? Would be best to add a minimal representative sample case with numeric data and the final output. — Divakar
– Divakar, Commented Jun 6, 2016 at 17:50

Rahul K P · Accepted Answer · 2016-06-06 17:52:16Z

2

You can implement this with numpy.dot

In [1]: import numpy as np
In [2]: a
Out[2]: array([1, 2])
In [3]: b
Out[3]: 
array([[1, 1],
       [1, 1]])
In [4]: np.dot(a,b)
Out[4]: array([3, 3])

Here is more Reference about numpy.dot

edited Jun 6, 2016 at 17:52

answered Jun 6, 2016 at 17:49

Rahul K P

16.2k4 gold badges40 silver badges56 bronze badges

Sign up to request clarification or add additional context in comments.

7 Comments

Chrispresso Over a year ago

I was thinking the same thing. The only issue is if @hugodecasta has a weight of unequal dimensions, i.e. he has 3 outputs and 2 inputs

hugodecasta Over a year ago

For same input/output shapes it work but for like 3/2 it fail as expected but i thinck it should be easy to fix that

Rahul K P Over a year ago

The rule already implemented in numpy.dot. We don't want to worry about that.

hugodecasta Over a year ago

Yes but the problem is still here [1,2,3] DOT [ [1,1,1], [1,1,1] ] Does not wrok... because the two vectors are not aligned

hugodecasta Over a year ago

My miss sory Instead of using [1,2,3] DOT [ [1,1,1], [1,1,1] ] we have to use [ [1,1,1], [1,1,1] ] DOT [1,2,3]

|

Collectives™ on Stack Overflow

Python - cross array multiplication

1 Answer 1

7 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

7 Comments

Your Answer

Sign up or log in

Post as a guest

Related