28 Nov
2019
28 Nov
'19
2:24 p.m.
Dear Petia!
My approach is the same as Andrew's but
nevertheless it takes time to
connect a complex network having multiple layers and
feedback/feedforward connectivity even though I have prepared and saved
all parameters in advance, esspecially if I want to have sparse
connectivity.
Of course it takes time to re-instantiate a network completely to do
repeated experiments on pristine networks, but so would saving and
loading. In fact, all our attempts to implement this functionality
showed that it actually often takes *more* time to restore a network
from disk than it would to just re-running the script that created it.
The main reason for this is that memory and processors are lightning
fast compared to disks.
The other reason for not having any checkpointing functionality and
removing ResetNetwork in NEST 3 is that is just crazy complicated to get
this right. There are spike buffers in neurons and inside the NEST
kernel, modified default values, random number generator states,
internal flags indicating simulation phases, possibly open files, and so
on and so forth. Because it is almost impossible to implement this in a
way that is both complete and correct, we rather not provide it.
I was also wandering is it possible to have sparse
connectivity matrix W
to be used in the following manner: connect(pop1, pop2, "all_to_all",
W). Now I am generating full matrix with numerous zero elements but if I
want to have dynamic synapses, initial zero weights might result in
non-zero one after some time.
You have to create all connections from the start if you want them to
exhibit dynamic weight changes during the simulation. The only exception
to this rule is structural plasticity:
https://www.fz-juelich.de/ias/jsc/EN/Expertise/SimLab/slns/research/structu…
Best regards,
Jochen!
On Wednesday, November 27, 2019, 4:42:32 PM GMT+2,
Simon Brodeur
<simon.brodeur(a)usherbrooke.ca> wrote:
Hi Andrew,
Is it because the network takes a long time to build?
Are you working with very large networks that need to be spread on
multiple machines?
I am personally building networks with complex topologies, where
creating the synaptic connections requires a lot of time since I do
random sampling with constraint satisfaction directly in Python.
I also need to compute per-synapse approximations of axonal delays,
dendritic tree attenuations and much more. I do build the network once
and have written some Python code that allow to pickle the necessary
information (e.g. parameters of the neuron models, synaptic weights) to
instantiate faster the network in NEST when I want to perform
simulations. But that is just a custom solution, not general to any network.
Cordially,
Simon
On Tue, 2019-11-26 at 12:39 +0100, alehr wrote:
--
Dr. Jochen Martin Eppler
Phone: +49(2461)61-96653
----------------------------------
Simulation Laboratory Neuroscience
Jülich Supercomputing Centre
Institute for Advanced Simulation
------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------
Forschungszentrum Juelich GmbH
52425 Juelich
Sitz der Gesellschaft: Juelich
Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498
Vorsitzender des Aufsichtsrats: MinDir Volker Rieke
Geschaeftsfuehrung: Prof. Dr.-Ing. Wolfgang Marquardt (Vorsitzender),
Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt,
Prof. Dr. Sebastian M. Schmidt
------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------
Dear NEST developer,
I am wondering if my inquiry from November 11th has been looked at.
Thanks,
Andrew Lehr
On Mon, Nov 11, 2019 at 1:52 PM alehr <alehr(a)mun.ca
<mailto:alehr@mun.ca>> wrote:
--
___________________________________________________
*Simon Brodeur*
/Étudiant au doctorat/
Université de Sherbrooke
Département génie électrique et génie informatique
Laboratoire NECOTIS, C1-3036
Tél. : (819) 821-8000 poste 62187
Courriel: Simon.Brodeur(a)USherbrooke.ca <mailto:Simon.Brodeur@USherbrooke.ca>
___________________________________________________
_______________________________________________
NEST Users mailing list -- users(a)nest-simulator.org
<mailto:users@nest-simulator.org>
To unsubscribe send an email to users-leave(a)nest-simulator.org
<mailto:users-leave@nest-simulator.org>
_______________________________________________
NEST Users mailing list -- users(a)nest-simulator.org
To unsubscribe send an email to users-leave(a)nest-simulator.org
Dear NEST developer,
I would like to initialize a network with neurons and connections and
then run many simulations with it. It would be great if I could build
the network one time and then deepcopy (or something similar) for
each simulation. Is something like this possible?
Thanks,
Andrew Lehr
_______________________________________________
NEST Users mailing list --users(a)nest-simulator.org
<mailto:users@nest-simulator.org>
To unsubscribe send an email tousers-leave(a)nest-simulator.org
<mailto:users-leave@nest-simulator.org>