MATLAB Central Newsreader - How to generate randomized weights for network topology?

How to generate randomized weights for network topology?

Marat — Wed, 20 May 2009 14:00:20 -0400

Hi all!
I need some help in generating values based on co-ordinates. The values are meant as weights for network. For e.g. for nodes 1->7 it is 23, then for 7->1 should also be 23. Similarly for a topology of size 20, I want to generate the weights for several nodes. Of course, some of the links can have 0s. Is this possible in matlab? Can someone please point out?

Thanks

Re: How to generate randomized weights for network topology?

Greg Heath — Thu, 21 May 2009 08:15:51 -0400

On May 20, 10:00 am, "Marat " <nemesis_...@yahoo.com> wrote:
> Hi all!
> I need some help in generating values based on co-ordinates. The values a=
re meant as weights for network. For e.g. for nodes 1->7 it is 23, then for=
7->1 should also be 23. Similarly for a topology of size 20,

I dom't understand what you mean. Your "example" makes no sense to me.

>I want to generate the weights for several nodes. Of course, some of the l=
inks > can have 0s. Is this possible in matlab? Can someone please point ou=
t?

Which neural networks? Some (e.g., NEWFF) generate their own random
initial weights.

Need more info.

Greg

Re: How to generate randomized weights for network topology?

Greg Heath — Thu, 21 May 2009 08:17:02 -0400

On May 20, 10:00 am, "Marat " <nemesis_...@yahoo.com> wrote:
> Hi all!
> I need some help in generating values based on co-ordinates. The values a=
re meant as weights for network. For e.g. for nodes 1->7 it is 23, then for=
7->1 should also be 23. Similarly for a topology of size 20, I want to gen=
erate the weights for several nodes. Of course, some of the links can have =
0s. Is this possible in matlab? Can someone please point out?
>
> Thanks

doc init
help init

Hope this helps.

Greg

Re: How to generate randomized weights for network topology?

Marat — Fri, 22 May 2009 02:48:01 -0400

My apologies for not being clear. I have a network topology of 3 nodes, and I need to have weights for the links between nodes thats is of this pattern:

0 10 4
10 0 10
4 10 0

This represents the weights between nodes, say between nodes 1 and 2 is 20, and since it is bi-directional the weights between nodes 2 and 1 is also 20. I need help in generating such weights for larger number of nodes.

Thanks you for your time.

Re: How to generate randomized weights for network topology?

Greg Heath — Fri, 22 May 2009 10:23:39 -0400

On May 21, 10:48 pm, "Marat " <nemesis_...@yahoo.com> wrote:
> My apologies for not being clear. I have a network topology of 3 nodes, and I need to have weights for the links between nodes thats is of this pattern:
>
> 0 10 4
> 10 0 10
> 4 10 0
>
> This represents the weights between nodes, say between nodes 1 and 2 is 20, and since it is bi-directional the weights between nodes 2 and 1 is also 20. I need help in generating such weights for larger number of nodes.
>
> Thanks you for your time.

You are still not clear.

What type of net do you have?

For example, an I-H-O FF MLP can be created using
net = newff(p,t,H);
Modifiable weights are contained in net.IW,net.LW and net.b.

Still need more info.

Greg

Re: How to generate randomized weights for network topology?

Marat — Fri, 22 May 2009 12:48:02 -0400

It is a static network obtained using a separate program, not matlab. The computer network topology is already present. It does not denote neural networks or any other similar types.

I am just looking for a way to generate the weights representing Quality of service values.

Thanks again.

Re: How to generate randomized weights for network topology?

Alan B — Fri, 22 May 2009 14:32:01 -0400

"Marat " <nemesis_810@yahoo.com> wrote in message <gv66u2$q8r$1@fred.mathworks.com>...
> It is a static network obtained using a separate program, not matlab. The computer network topology is already present. It does not denote neural networks or any other similar types.
>
> I am just looking for a way to generate the weights representing Quality of service values.
>
> Thanks again.

If I understand correctly, you can do this:

a = floor(rand(3)*10)
a = triu(a) + triu(a,1)'