Re: [moose-devel] An error in testHSolve.py

[Subhasis R. <ray...@gm...>]

Dear Saeed,
  (CC to Upi, Dilawar, Vivek, the developers involved with components
related to this issue)

On Fri, Jul 4, 2014 at 5:09 AM, saeed <sa...@gn...> wrote:

>     Application crashes with Assertion failure in SpikeRingBuffer.cpp:63
>     and actually I just put the warning to maybe guide to the problem.
>     My change is not about the dt, after just a little increase in
>     simulation time (from 0.01 to 0.1) I see this problem.


Sorry for overlooking this point in my hurry. The documentation in
SpikeRingBuffer.h says:

/**
 * This ring buffer handles incoming spikes. It spans an interval equal to
 * the longest arrival delay. When a spike event is notified it puts it into
 * the slot defined by the arrival time. It just adds the weight onto the
 * existing value in this slot, assuming a linear summation of the
 * weights of coincident inputs.
 */

It was designed to simplify storage of spike events coming with a specified
delay period and each slot in the buffer corresponds to the time step in
which it should be taken into account for computing the synaptic
conductance. Given that axonal delays can be of the order of 100 ms on the
slower side while dt is usually of the order of 100 us, I think we have a
valid design issue here. Thanks for bringing this up.

I am wondering, maybe SynChan is not working properly. I know that the
>     HSolve is still under control, but in other side I finished the GPU
>     implementation and for accuracy test, I need to make the Moose work.
>

Since SynChan is outside the purview of HSolve, you should not be hindered
by this particular case. Also, hsolve seems to work for simple cases (like
testHsolve  and Rallpacks), though it has not been tested rigorously for
complex neuron models. So you can try to plugin your code and check if the
GPU code works for the simple cases first while Dilawar ties up the loose
ends.

Best,
  Subhasis


> On 02-07-2014 15:25, Subhasis Ray wrote:
> > Dear Saeed,
> >    You mean simulation timestep (dt) rather than simulation time. The
> > warning is harmless. The SpikeRingBuffer is a ring-buffer for
> > temporarily storing spikes and the size of the buffer is is
> > SynChan.bufferTime/dt. So if you make dt small without reducing
> > bufferTime, it avoids increasing the number of entries beyond the
> > predefined maximum.
> >    That aside, the HSolve class has not yet been fully ported to
> > async13 branch and I encourage you to modify the testHSolve script to
> > test this.
> >
> > Best,
> >    Subhasis
> >
> > On 7/2/14, Saeed Shariati <sd....@gm...> wrote:
> >> Dear all,
> >>
> >> In the Demo/snippents/testHSolve.py file, when the simulation time
> >> increases (for example 0.1 instead of 0.01), application exits with the
> >> below error:
> >>
> >> *Warning: SpikeRingBuffer: bin number exceeds limit: spikeTime =
> 0.047002,
> >> currtime=  0, dt = 0.0001, bin = 470 >=  128, terminating*
> >> *python: SpikeRingBuffer.cpp:63: void SpikeRingBuffer::addSpike(double,
> >> double): Assertion `0' failed.*
> >>
> >> I've compiled with BUILD=debug.
> >> Any clue would be appreciated.
> >>
> >> Thank you,
> >> Saeed
> >>
> ------------------------------------------------------------------------------
> >> Open source business process management suite built on Java and Eclipse
> >> Turn processes into business applications with Bonita BPM Community
> Edition
> >> Quickly connect people, data, and systems into organized workflows
> >> Winner of BOSSIE, CODIE, OW2 and Gartner awards
> >> http://p.sf.net/sfu/Bonitasoft
> >> _______________________________________________
> >> moose-devel mailing list
> >> moo...@li...
> >> https://lists.sourceforge.net/lists/listinfo/moose-devel
> >>
> >
>
>
> ------------------------------------------------------------------------------
> Open source business process management suite built on Java and Eclipse
> Turn processes into business applications with Bonita BPM Community Edition
> Quickly connect people, data, and systems into organized workflows
> Winner of BOSSIE, CODIE, OW2 and Gartner awards
> http://p.sf.net/sfu/Bonitasoft
> _______________________________________________
> moose-devel mailing list
> moo...@li...
> https://lists.sourceforge.net/lists/listinfo/moose-devel
>