Benchmarking the working pipeline

[albertoarturovergani]

Create a notebook which computes the time consumption of substructures of code that are essential in building and running the SNNs simulations.

The candidate notebook should have the following substructures (cell or set of cells):

• [ ] get the benchmark intuition from http://neuralensemble.org/docs/PyNN/examples/VAbenchmarks.html

• [ ] select the proper simulator as backend of PyNN, e.g., import pyNN.spiNNaker as sim (server at https://spinn-20.cs.man.ac.uk/hub/home) import pyNN.nest as sim (info at http://neuralensemble.org/docs/PyNN/backends/NEST.html)

• [ ] building step of a simple network without topology (aka no projections), e.g.,

  • define the input layer (n neurons)
  • define the excitatory layer (n neurons)
  • define the inhibitory layer (n/5 neurons)
  • select the dedicated neuron model and parameters

• [ ] projections step for the network, e.g.,

  • define the connectivity matrix assigning correspondence between neurons,
  • define weights and proper delays
  • the connectivity should follow a given geometrical distribution (exponential families) as well as standard cases (All2All, One2One, random, etc)
  • resulting as a numpy list that PyNN should read within the given constructor (FromListConnector) http://neuralensemble.org/docs/PyNN/_modules/pyNN/connectors.html#FromListConnector

[laurentperrinet]

I just pushed in the benchmark folder a set of notebooks that already worked for doing a benchmark of the pipeline. We should:

1. test that it still works with the current version of the VisionSNN scripts (+ on pyNNN / spinnaker),
2. re-run them when we integrate the novel parts in the code...