[PD-dev] Segfaults with FLEXT code

[B. Bogart]

Hey Thomas,

Thanks for the hints, I put in a few FLEXT_ASSERTS and I wrapped the array
references with an if statement to make sure the values are in the valid
range. Still no luck.

I wish I could send a patch that would explain the behavior, but the reason
why I'm doing it as an external is that I think it would be too complicated
to do as a patch, at least it would be just as time consuming. If you refer
to the python code I sent you way back (attached again) thats the exact
behavior the external requires. I'd love to get you more informarion but I
only half-understand the algo myself. I'll give it a try...

Particle Swarm Optimizer:

Very similar to a Genetic Algo. You have a population of solutions to a
problem, randomly generated. Then through an iterative process you try and
steer the population towards the correct solution. In the case of the PSO
The particles are attracted to the member with the best solution. So the
whole population accelerates towards the best solution. The theory goes that
the particles accelerating towards the best solution will find the solution
and all the paricles will end up on the best solution. The python program
just minimizes the distance from the particles initial position to the
center. My external will translate the position of the "center" to match up
with a target. this way the particles will all gravitate towards this
target.

Does that help?

I fixed the rand() issue, not sure what I was thinking since I was already
dividing by RAND_MAX!

Since its not the array indexing is there anything else that looks suspect
in your eyes? I've been starring at it for days, along side the C version
that works fine, and I just can't spot any problems. Since the problem is in
the iterate function I really don't know how to approach trouble-shooting
it, because any change I make will change the behavior of the pso and
whatever is causing the fault may or may not realize itself. The system is
supposed to be emergent by definition!

What is the standard way to deal with these problems?

I did not realize I could omit " this-> ", but I think i'll leave it all in
place until I can get rid of this fault!

Thanks for your help.

Ben

----- Original Message -----
From: "Thomas Grill" <t.grill at gmx.net>
To: "B. Bogart" <ben at ekran.org>; <pd-dev at iem.at>
Sent: Tuesday, August 05, 2003 5:39 AM
Subject: Re: [PD-dev] Segfaults with FLEXT code


> Hi Ben,
>
> it would be great if you could send a patch that demonstrates some
intended
> behavior of your external..... otherwise i can't really figure out what it
> does.... Anyway, having a short look at the code i spotted some things:
>
> 1) RAND_MAX is defined in stdlib.h to represent the largest number rand()
> will ever deliver. This means that it is guaranteed that 0 <=
> rand()/RAND_MAX <= 1. If you redefine RAND_MAX in your code to some
> different value this is no longer true. rand() doesn't know about the
> redefinition! So, if you want your random numbers to be in a range from a
to
> b then you should use something like  r = a+(float)rand()/RAND_MAX*(b-a).
If
> the granularity of rand() is too large for your application you might also
> want to check out better algorithms (see for example
> http://www.library.cornell.edu/nr/bookcpdf.html , chapter 7).
>
> 2) You have some tricky array indexing in your code, which could be a
> potential cause of trouble (e.g. because of a simple typo). Maybe you
would
> want to insert some assertion statements (or other means) into your code
to
> verify that the indexing is correct.
>
> for example:
> int ix = p*hoodsize+n;
> FLEXT_ASSERT(ix < hoodsize*popsize); // check if index is within the array
> bounds
> neighbors [ix] = p-1+n;
>
> 3) Are you aware of the fact that within class members the this-pointer
can
> be omitted?  This means that e.g. "popsize" means exactly the same as
> "this->popsize" in any member function of the pso class. On the other hand
> explicit usage of this can make things clearer, of course.
>
> best greetings,
> Thomas
>
>
>
> ----- Original Message -----
> From: "B. Bogart" <ben at ekran.org>
> To: <pd-dev at iem.at>
> Sent: Tuesday, August 05, 2003 5:10 AM
> Subject: [PD-dev] Segfaults with FLEXT code
>
>
> > Hello all,
> >
> > I'm having trouble with segfaults with my FLEXT external, I'm new to
flext
> > and C++ and I really don't have an idea how to narrow it down. I've got
it
> > down to two lines that commented out remove the segfault, but I can't
find
> > anything wrong with those lines.
> >
> > I've attached my flext class (pso.cpp) and the SDL C program I'm
> > re-implimenting. The SDL program has no segfault problems, and most of
the
> > code is copy-pasted with only variable and array subscript changes...
> >
> > Any help would be greatly appreciated.
> >
> > Thanks
> > Ben
> >
>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: particleswarm.py.txt
URL: <http://lists.puredata.info/pipermail/pd-dev/attachments/20030807/eb766445/attachment.txt>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pso.cpp
Type: application/octet-stream
Size: 9705 bytes
Desc: not available
URL: <http://lists.puredata.info/pipermail/pd-dev/attachments/20030807/eb766445/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: sdl_orig.c
Type: application/octet-stream
Size: 3734 bytes
Desc: not available
URL: <http://lists.puredata.info/pipermail/pd-dev/attachments/20030807/eb766445/attachment-0001.obj>