Cortical microcircuit simulation: PyNN version

Stored for easy access for people within the organisation

Contributors: Sacha van Albada ([email protected]) Maximilian Schmidt Jannis Schücker Andrew Rowley Alan Stokes

This is an implementation of the multi-layer microcircuit model of early sensory cortex published by Potjans and Diesmann (2014) The cell-type specific cortical microcircuit: relating structure and activity in a full-scale spiking network model. Cerebral Cortex 24 (3): 785-806, doi:10.1093/cercor/bhs358

It has been run on three different back-ends: NEST, SpiNNaker, and the ESS (emulator of HMF)

Instructions

1. Ensure you have the desired back-end.

   For SpiNNaker see https://spinnakermanchester.github.io/latest/spynnaker.html

   For NEST see http://www.nest-initiative.org/index.php/Software:Download and to enable full-scale simulation, compile it with MPI support (use the --with-mpi option when configuring) according to the instructions on http://www.nest-initiative.org/index.php/Software:Installation

2. Install PyNN according to the instructions on http://neuralensemble.org/docs/PyNN/installation.html

3. Run the simulation by typing python run_microcircuit.py <simulator> in your terminal in the folder containing this file, where <simulator> is one of nest or spinnaker (by default spinnaker is selected). There are several potential arguments which can be seen by typing python run_microcircuit.py <simulator> -h. A few useful ones include:

   • --sim_duration - The simulation duration in milliseconds (default 1000)
   • --output_path - Where output files should be written (default results)

4. Output files and basic analysis:

   • Spikes are written to .txt files containing IDs of the recorded neurons and corresponding spike times in ms. Separate files are written out for each population and virtual process. File names are formed as 'spikes'+ layer + population + MPI process + .txt

   • Voltages are written to .dat files containing GIDs, times in ms, and the corresponding membrane potentials in mV. File names are formed as voltmeter label + layer index + population index + spike detector GID + virtual process + .dat

   • If 'plot_spiking_activity' is set to True, a raster plot and bar plot of the firing rates are created and saved as 'spiking_activity.png'

This simulation is part of the Spynnaker integration tests so is tested daily. The tests use python 3.12 and the latest possible version of each dependency unless restricted by https://github.com/SpiNNakerManchester/sPyNNaker/blob/master/requirements.txt

Known issues:

• At least with PyNN 0.7.5 and NEST revision 10711, ConnectWithoutMultapses works correctly on a single process, but not with multiple MPI processes.

• When saving connections to file, ensure that pyNN does not create problems with single or nonexistent connections, for instance by adjusting lib/python2.6/site-packages/pyNN/nest/init.py from line 365 as follows:

  if numpy.size(lines) != 0:
      if numpy.shape(numpy.shape(lines))[0] == 1:
          lines = numpy.array([lines])
          lines[:,2] *= 0.001
          if compatible_output:
              lines[:,0] = self.pre.id_to_index(lines[:,0])
              lines[:,1] = self.post.id_to_index(lines[:,1])
      file.write(lines, {'pre' : self.pre.label, 'post' : self.post.label})
  

• To use saveConnections in parallel simulations, additionally ensure that pyNN does not cause a race condition where the directory is created by one process between the if statement and makedirs on another process: In lib/python2.6/site-packages/pyNN/recording/files.py for instance replace

  os.makedirs(dir)
  

  by

  try:
      os.makedirs(dir)
  except OSError, e:
      if e.errno != 17:
          raise
      pass
  

Reinstall pyNN after making these adjustments, so that they take effect in your pyNN installation directory.