Skip to content

Latest commit

 

History

History
53 lines (38 loc) · 3.92 KB

JupyterHub_Setup.md

File metadata and controls

53 lines (38 loc) · 3.92 KB
Raw

Using JupyterHub and SpiNNaker

The Jupyter Notebook is a tool where code can be executed in cells within a browser interface. The code is executed by the Kernel which receives instructions from the Jupyter Notebook frontend.

JupyterHub is a tool allowing multiple users to spawn instances of a Jupyter Notebook server installed on a local machine, so that other people can run Jupyter notebooks as if they were on the local machine.

This means that in demonstrations, users can access a SpiNNaker installation and use PyNN from any device, as long as they have a user account on the serving machine.

Installing JupyterHub

Make sure all services are installed into the system as root instead of into a user directory so that all users can access it.

  • Install Jupyter using pip3 and sudo to ensure that Jupyter runs from a Python3 installation. Do not use Anaconda as it creates a Python installation in the local user directory.
  • Install JupyterHub - requires Python3.4 or greater.
  • Create a jupyter_config.py file and add your user as admin
  • Start JupyterHub by running sudo jupyterhub from the folder which contains the config file.

By nature, root can control the amount of access users have. For more security, use a HTTPS connection

Using sPyNNaker8

  • Make sure the SpiNNaker files are in a system directory (e.g. /usr/share/) and that the correct system environmental variables are setup in /etc/profile.d or /etc/profile/
  • Move the .spynnaker.cfg file to /etc/xdg instead of a user directory so that all users read that instead of creating their own.

Custom Kernels

A custom kernel has been created to provide additional commands to be used in the notebook. This works by intercepting the code sent by the notebook. The kernel runs python2.7 so it's compatible with sPyNNaker.

Current commands include:

  • ##printmysession prints a record of all code run in the notebook since restart/reset, to provide easy debug.
  • ##printmyout prints a record of all output sent to stdout or stderr from each cell (if recording is on).
  • ##printmysessionout etc. does both of these in one, to provide easy debug.
  • ##recordoutput[on/off] toggles between recording output or not. Note when recording output, images are disabled and output is not shown linearly to the user. Default off.
  • ##rerun x y executes all the code in cells between cell x and cell y, so you can for example rerun a set of setup cells without executing them individually. ##rerun x just executes cell x.

More commands can be easily added.

The kernel (which runs python2.7) consists of:

  • The .py file which should be placed in the python2.7/site-packages folder
  • The kernel folder containing the kernel.json, which should be placed in either /usr/local/share/jupyter/kernels/ or /usr/share/jupyter/kernels depending on your system.

The kernel can then be used in a Notebook by selecting it from the dropdown.

Using JupyterHub

  • Create a local user for each person:
    • sudo useradd -d /home/[username]/ -m [username]
    • sudo passwd [username]
  • Allow the connection through port 8000 in the firewall (note this is reset on reboot):
    • sudo iptables -A INPUT -p tcp --dport 8000 -j ACCEPT
  • Start the server (see above). Anyone can access the server and login as a local user from http://[your ip]:8000

Other notes

matplotlib can be used to plot inline graphs in a notebook. Ensure that it is higher than version 1.3. If TKinter is throwing up display problems, you need to change the matplotlib backend to Agg in the config files:

  • /etc/matplotlibrc
  • /usr/lib64/python2.7/site-packages/matplotlib......../matplotlib/mpl-data/matplotlibrc