latest

Installation

Local machine

On a local machine not connected to a Loihi host, you can use any version of Python that has pip.

git clone https://github.com/nengo/nengo-loihi.git
pip install -e nengo-loihi

pip will do its best to install Nengo Loihi’s requirements. If anything goes wrong during this process, it is likely related to installing NumPy. Follow our NumPy install instructions, then try again.

INRC/Superhost

These steps will take you through setting up a Python environment for running Nengo Loihi, as well as for running models using the NxSDK directly.

Note, you must use Python 3.5.2 and NumPy 1.14.3 when working with NxSDK. The easiest way to satisfy those constraints is to use Miniconda to set up an isolated environment for running Loihi models.

  1. Ensure that conda is available.

    To see if it is available, run

    conda -V
    

    If conda is available, the conda version should be printed to the console.

    If it is not available:

    1. Ask your superhost administrator if conda is installed. If it is, you need to add the bin directory of the conda installation to your path.

      export PATH="/path/to/conda/bin:$PATH"
      

      Running this once will change your path for the current session. Adding it to a shell configuration file (e.g., ~/.profile, ~/.bashrc) will change your path for all future terminal sessions.

    2. If conda is not installed, install Miniconda.

      wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh
      bash miniconda.sh
      

      Follow the prompts to set up Miniconda as desired.

  2. Create a new conda environment. Note, you must use Python 3.5.2 when working with NxSDK.

    conda create --name loihi python=3.5.2
    
  3. Activate your new environment.

    source activate loihi
    

    Sometimes the environment can have issues when first created. Before continuing, run which pip and ensure that the path to pip is in your conda environment.

    Note

    You will need to run source activate loihi every time you log onto the superhost.

  4. Install NumPy and Cython with conda. Note, you must use NumPy 1.14.3 when working with NxSDK.

    conda install numpy=1.14.3 cython
    

    The NumPy provided by conda is usually faster than those installed by other means.

  5. Copy the latest NxSDK release to your current directory.

    Note

    The location of NxSDK may have changed. Refer to Intel’s documentation to be sure. The most recent release and NxSDK location are current as of September 2018.

    If you are logged into INRC:

    cp /nfs/ncl/releases/0.7/nxsdk-0.7.tar.gz .
    

    If you are setting up a non-INRC superhost:

    scp <inrc-host>:/nfs/ncl/releases/0.7/nxsdk-0.7.tar.gz .
    
  6. Install NxSDK.

    pip install nxsdk-0.7.tar.gz
    
  7. Install Nengo Loihi.

    git clone https://github.com/nengo/nengo-loihi.git
    pip install -e nengo-loihi
    

    pip will install other requirements like Nengo automatically.

  8. Test that both packages installed correctly.

    Start Python by running the python command. If everything is installed correctly, you should be able to import nxsdk and nengo_loihi.

    Python 3.5.2 |Anaconda, Inc.| (default, May 13 2018, 21:12:35)
    [GCC 7.2.0] on linux
    Type "help", "copyright", "credits" or "license" for more information.
    >>> import nxsdk
    >>> import nengo_loihi