PyNN microcircuit

[neworderofjamie]

I'm trying to get the PyNN version of the microcircuit working and am really struggling to find the right combination of NEST and PyNN version. The readme says to use Nest 2.4.1 but I have failed to find a version of PyNN that is suitable:

• PyNN 0.8.0 (installed from pip) fails with AttributeError: 'module' object has no attribute 'FixedTotalNumberConnector'
• PyNN 0.8.1 (installed from pip) fails with pynestkernel.NESTError: UnknownModelName in SetDefaults_l_D: /static_synapse_lbl is not a known model name.

Could you update the README so it recommends the correct versions? Additionally, what are the scaling values required to match the version in the paper (80x10^3 neurons, 0.3x10^9 synapses)? In the PyNest version they're both set to 0.1 and in the PyNN version K_scaling = 0.5 and N_scaling = 1.0.

[albada]

Thanks for your comment, it does look like the README is outdated. Looking at the latest commits, it seems like it should work with NEST 2.12. Could you try that? The version of the model matching that in the paper is obtained with K_scaling = 1.0, N_scaling = 1.0.

[neworderofjamie]

No problem! Thanks for the quick response. I will have a go with NEST 2.12 today - what version of PyNN does that work with?

[albada]

According to the commit message for the addition of the PyNN code to this repository, it should work with PyNN 0.8 (I don't know in more detail, but I guess the additional detail doesn't matter)

[neworderofjamie]

Hi Sasha,

Testing with Nest 2.12.0:

• PyNN 0.8.0 doesn't work because the FixedNumberConnector seems to be missing
• PyNN 0.8.1 fails with:

  File "microcircuit.py", line 19, in <module>
    exec('import pyNN.%s as sim' % simulator)
  File "<string>", line 1, in <module>
  File "/home/j/jk/jk421/nest_2_12_0/local/lib/python2.7/site-packages/pyNN/nest/__init__.py", line 38, in <module>
    from pyNN.nest.recording import *
  File "/home/j/jk/jk421/nest_2_12_0/local/lib/python2.7/site-packages/pyNN/nest/recording.py", line 112, in <module>
    class Multimeter(RecordingDevice):
  File "/home/j/jk/jk421/nest_2_12_0/local/lib/python2.7/site-packages/pyNN/nest/recording.py", line 114, in Multimeter
    _nest_connect = nest.ConvergentConnect
AttributeError: 'module' object has no attribute 'ConvergentConnect'
(nest_2_12_0) jk421@inf900801:~/PotjansDiesmann2014-master/PyNN$ pip install pynn==0.8.2

• PyNN 0.8.2 fails with

Traceback (most recent call last):
  File "microcircuit.py", line 26, in <module>
    n.setup(sim)
  File "/home/j/jk/jk421/PotjansDiesmann2014-master/PyNN/network.py", line 100, in setup
    syn = sim.StaticSynapse(weight=self.w_ext)
  File "/home/j/jk/jk421/nest_2_12_0/local/lib/python2.7/site-packages/pyNN/models.py", line 105, in __init__
    raise e
KeyError: 'min_delay'

• PyNN 0.8.3 fails with

Traceback (most recent call last):
  File "microcircuit.py", line 26, in <module>
    n.setup(sim)
  File "/home/j/jk/jk421/PotjansDiesmann2014-master/PyNN/network.py", line 128, in setup
    d_mean[source_pop], d_sd[source_pop])
  File "/home/j/jk/jk421/PotjansDiesmann2014-master/PyNN/connectivity.py", line 22, in FixedTotalNumberConnect
    proj = sim.Projection(pop1, pop2, conn, syn, receptor_type=conn_type)
  File "/home/j/jk/jk421/nest_2_12_0/local/lib/python2.7/site-packages/pyNN/nest/projections.py", line 60, in __init__
    connector.connect(self)
  File "/home/j/jk/jk421/nest_2_12_0/local/lib/python2.7/site-packages/pyNN/connectors.py", line 1029, in connect
    connections[target_index].append(source_index)
TypeError: only integer scalar arrays can be converted to a scalar index

So I'm a little unsure what to do...

[pgleeson]

@neworderofjamie Have a look at the logs here: https://travis-ci.org/OpenSourceBrain/PotjansDiesmann2014/builds/316992950#.

These should show the details of the various packages which need to be installed to get the PyNN + NEST example working.

[pgleeson]

Also, a Docker file which builds the full environment required is here: https://github.com/OpenSourceBrain/neuralensemble-docker/blob/python2/osb/Dockerfile

[neworderofjamie]

Hi Padraig,

Thanks for that - your test looks like it checks out master from the PyNN github and doing that works for me. It also seems to work with 0.9.2 so, to save other people this hassle, could you update the readme to say the release version it works with - master on github seems a bit risky/something of a moving target :)

Cheers

Jamie

[pgleeson]

@neworderofjamie Good point, the Readme(s) certainly need updating, and there's alos an open PR from @apdavison which needs to be checked... I plan to have a good look over this in the next week or so.