{5} Assigned, Active Tickets by Owner (Full Description) (10 matches)

I notice that the pyNN TsodyksMarkramSynapse? uses the nest tsodyks_synapse ... I don't think this is the right one ... because you don't even consider the parameter "tau_psc_", which is the time constant of postsyn current, that is the tsodyks et al 2000 model includes modeling of that. If you want the typical Markram+Tsodyks 1998 UDF model, the tau_psc part is of the tsodyks_synapse should not be there ... so I implemented the nest markram_synapse according to Markram+Tsodyks 1998 ... its in the developer directory ... Since there is no exposure of tau_psc, I think the PyNN model intends this one, does it not?

Following on from ticket:111, since reset() does not change the list of things to record, but it may be desirable to record different things on different runs, we also need a mechanism for resetting the list of things to be recorded, e.g. record(None).

With pyNN.neuron, when the TsodyksMarkramMechanism?(TMM) is used targeting exponential synapses, the synaptic conductance becomes "nan" in specific setups: If the TMM recovery time constant "tau_rec" is equal to the decay time constant of the target synapse ( "tau_syn_E" or "tau_syn_I"), gsyn becomes "nan" and it seems as if there was no synaptic connection at all in the network.

Using the same values for both time constants is probably not very biological, but may be useful when implementing saturating synapses. The facilitation time constant "tau_facil" has no effect on that behavior.

See an example below, reproducing this errors. This is the tsodyksmarkram.py example script, only the time constants were adapted accordingly.

import numpy
from pyNN.utility import get_script_args

simulator_name = get_script_args(1)[0]
exec("import pyNN.%s as sim" % simulator_name)

sim.setup(debug=True, quit_on_end=False)

spike_source = sim.Population(1, sim.SpikeSourceArray,
                              {'spike_times': numpy.arange(10, 100, 10)})

connector = sim.AllToAllConnector(weights=0.01, delays=0.5)

synapse_dynamics = {
    'static': None,
    'depressing': sim.SynapseDynamics(
        fast=sim.TsodyksMarkramMechanism(U=0.5, tau_rec=10.0, tau_facil=0.0)),
    'facilitating': sim.SynapseDynamics(
        fast=sim.TsodyksMarkramMechanism(U=0.04, tau_rec=10.0, tau_facil=1000.0)),
}

populations = {}
projections = {}
for label in 'static', 'depressing', 'facilitating':
    populations[label] = sim.Population(1, sim.IF_cond_exp, {'e_rev_I': -75, 'tau_syn_I':10.}, label=label)
    populations[label].record_v()
    if populations[label].can_record('gsyn'):
        populations[label].record_gsyn()
    projections[label] = sim.Projection(spike_source, populations[label], connector,
                                        target='inhibitory',
                                        synapse_dynamics=synapse_dynamics[label])

spike_source.record()

sim.run(200.0)

for label,p in populations.items():
    p.print_v("Results/tsodyksmarkram_%s_%s.v" % (label, simulator_name))
    if populations[label].can_record('gsyn'):
        p.print_gsyn("Results/tsodyksmarkram_%s_%s.gsyn" % (label, simulator_name))

print spike_source.getSpikes()

sim.end()

Currently, I tend to do:

  import sys
  simulator = sys.argv[1]
  exec("import pyNN.%s as sim" % simulator)

to let the simulator be specified on the command line.

To reduce this to a single statement, and add error checking, it would be useful to encapsulate it in a function in common, and use optparse from the standard library to handle passing additional arguments and options.

Some simulators allow writing variables other than spike times and membrane potential to file. To support these, we should have a single function and a single Population method for writing variables to file.

Also, python tends to use write(), rather than print(), for this kind of thing.

This should be done in a separate branch until we've made a release with the current API.