PyNN on SpiNNaker Installation Guide

Follow me on GitHub

This page describes an older version. Please see the latest version or home page

This guide details how to install the release version of the tools required to run simulations on SpiNNaker using the PyNN scripting language, if you have a local SpiNNaker machine. If you wish to run jobs on the million-core SpiNNaker machine in Manchester, then we suggest you use our Jupyter Notebook instructions, or submit jobs via the HBP Portal instead.

Installation

You must first install the Python Dependencies. Continue to the Virtualenv Installation, Central Installation, or User-only Installation to install the remaining requirements, depending on the needs of your working environment. If you are unsure what to choose, the option that will most likely work is the Virtualenv Installation.

Note: sPyNNakerExternalDevicesPlugin and sPyNNakerExtraModelsPlugin are no longer required. These extra modules have been merged in to the main sPyNNaker module. You will be asked to uninstall these modules below if you have previously installed them, but will note that there are no instructions for installing a newer version.

Please be aware due to an unexpected change to PyNN the released versions only work with PyNN version 0.9.0 to 0.9.4.

Please make sure you have installed PyNN version 0.9.4 before running the rest of these instructions.

Virtualenv Installation

Follow these instructions to install the dependencies in a virtual environment.

If you already have installed sPyNNaker previously (and the optional sPyNNakerExternalDevicesPlugin and/or sPyNNakerExtraModelsPlugin), you will need to uninstall it:

  1. Activate your virtualenv, <name>

    source <name>/bin/activate
    
  2. Uninstall the existing tools

    pip uninstall pyNN-SpiNNaker
    pip uninstall sPyNNaker
    pip uninstall sPyNNakerExternalDevicesPlugin
    pip uninstall sPyNNakerExtraModelsPlugin
    

To install the tools, first, activate your virtualenv, <name>:

source <name>/bin/activate
  1. Install Matplotlib:

    pip install matplotlib
    
  2. Install sPyNNaker8:

    pip install sPyNNaker8
    
  3. Install pyNN-SpiNNaker:

    python -m spynnaker8.setup_pynn
    

You can now configure your environment.

Central Installation

These instructions will install the required packages in a central location. If you are installing on Linux, you must have root access to do this (or prepend each command with sudo), and Windows users should run these commands from a command prompt run as Administrator (right-click the shortcut for the command prompt and select “Run as administrator”, but do not use the sudo in the commands below).

If you already have installed sPyNNaker previously (and the optional sPyNNakerExternalDevicesPlugin and/or sPyNNakerExtraModelsPlugin), you will need to uninstall it:

[sudo] pip uninstall pyNN-SpiNNaker
[sudo] pip uninstall sPyNNaker
[sudo] pip uninstall sPyNNakerExternalDevicesPlugin
[sudo] pip uninstall sPyNNakerExtraModelsPlugin
  1. Install Matplotlib:

    [sudo] pip install matplotlib
    
  2. Install sPyNNaker8:

    [sudo] pip install sPyNNaker8
    
  3. Install pyNN-SpiNNaker:

    [sudo] python -m spynnaker8.setup_pynn
    

You can now configure your environment.

User-only Installation

These instructions will install the required packages only for the current user (in your home directory). This can avoid issues where you don’t have root access to the machine you are installing on, or on a shared machine where the dependencies might otherwise conflict with those of other users.

If you already have installed sPyNNaker previously (and the optional sPyNNakerExternalDevicesPlugin and/or sPyNNakerExtraModelsPlugin), you will need to uninstall it:

pip uninstall pyNN-SpiNNaker
pip uninstall sPyNNaker
pip uninstall sPyNNakerExternalDevicesPlugin
pip uninstall sPyNNakerExtraModelsPlugin
  1. Install Matplotlib:

    pip install matplotlib --user
    
  2. Install sPyNNaker8:

    pip install sPyNNaker8 --user
    
  3. Install pyNN-SpiNNaker:

    python -m spynnaker8.setup_pynn
    

You can now configure your environment.

Configuration

When SpyNNaker is first called, if a configuration file is not found, it will create one in your home directory and exit. It is possible to ask SpyNNaker to do this before you run your first simulation as follows:

Run this small script

import pyNN.spiNNaker as sim
sim.setup()
sim.end()

Unless one already exists a new file called “.spynnaker.cfg” will be created in your home directory. You must edit this file to ensure that SpyNNaker can access your SpiNNaker machine. Upon opening this file, the part to alter will look like the following:

[Machine]
machineName = None
version = None

If you have a SpiNNaker board, then go to Local Board.

If you do not have a SpiNNaker board, then you have two options:

1) If you can directly access a local machine that uses spalloc (for example, you are in Manchester and wish to use the million-core machine), then you need to set the following parameters in the “.spynnaker.cfg” you just created (e.g.):

[Machine]
spalloc_server = spinnaker.cs.man.ac.uk
spalloc_user = user.name@email.address

where you edit spalloc_server if you are using a different spalloc_server; editing spalloc_user is helpful for administrators of the machine to contact you if there are any problems, which is why we suggest using an email address.

OR 2) To run in virtual mode, please follow the instructions in Instructions on how to use the different front ends in virtual mode and then go to Running some examples.

Local Board

Within the file, you should set machineName to the IP address or hostname of your SpiNNaker machine, and version to the version of your SpiNNaker board; this will almost certainly be “3” for a 4-chip board or “5” on a 48-chip board. The default ip address for a spinn-3 board is 192.168.240.253 and the default ip address for a spinn-5 board is 192.168.240.1.

Now go to Network Configuration.

Network Configuration

Go to the network settings for your computer and add or set an IPv4 entry with the following address for the adapter connected to the SpiNNaker board:

  1. IP address = 192.168.240.254
  2. sub-mask = 255.255.255.0
  3. default gateway = 0.0.0.0

Running some examples

  1. Download the examples:

    • PyNN 0.8 Examples are here (zip) or here (tar.gz)
  2. Extract the archive
  3. Go to the “examples” folder
  4. Run:

    python va_benchmark.py
    
  5. You will see the system go through a series of processes from partitioning, to placement, to routing and finally to loading and running.
  6. Once the example has finished, you should see a graph that will look something like this:

    VABenchmarkSpikes

If you get the output above, you have successfully installed your system.

Troubleshooting

  1. If when you run a script you see an error of the form:

    **TypeError: can't multiply sequence by non-int of type 'float'**
    

    this likely means that you have a newer version of PyNN that is not compatible with our released version of sPyNNaker8. To fix it, downgrade to PyNN version 0.9.4 by using:

     pip install pynn==0.9.4
    
  2. If on Windows you experience the error:

    **UnicodeDecodeError: 'ascii' codec can't decode byte 0xb0 in position 1: ordinal not in range(128)**
    

    Then to solve it, edit C:\Python27\Lib\mimetypes.py and add these codes:

    if sys.getdefaultencoding() != 'gbk':
        reload(sys)
        sys.setdefaultencoding('gbk')
    

    before the following line:

    default_encoding = sys.getdefaultencoding()
    
  3. In OSX, if experiencing the following tkinter error:

    _tkinter.TclError: no display name and no $DISPLAY environment variable
    

    it may be solved by setting the backend for matplotlib. This can be done by editing the matplotlibrc file in the current working directory to read to backend: TkAgg. This is usually found in $DEV/lib/python2.7/site-packages/matplotlib/mpl-data/matplotlibrc

    In a virtualenv, create a new file in the root directory .matplotlib/matplotlibrc that reads backend: TkAgg. (Sample matplotlibrc file)

    If you are still having issues, you may also need to install XQuartz.

  4. In OSX, if you have problems during the installation of the csa package (a dependency of sPyNNaker8; this problem cascades outwards) within your virtualenv, then use:

     MPLBACKEND=module://matplotlib.backends.backend_agg pip install sPyNNaker8 --user
    

    This overrides the matplotlib plotting backend just during the installation phase, which is sufficient to get a working installation if you are not actively using matplotlib to do immediate plotting of the data.

Other Links

Follow SpiNNaker Extensions to install extensions for building new neuron models.