jupyter - anjavdl/PHY517_AST443 GitHub Wiki

Jupyter notebooks are a great way to keep code organized and documented. Notebooks have individual cells, which can be in the kernel language (in this case, python), or in markdown (the language also used by github). A single cell can be run by hitting command-enter or shift-enter (the latter automatically advances to the next cell). A cheat-sheet for Markdown syntax can be found here. Note that markdown cells support LaTeX equations. To start jupyter, type

jupyter notebook

on your command line. This will open a browser window, from which you can navigate to the notebook of choice.

To exit jupyter, hit ctrl-c twice in the terminal from which you started it. Not doing so will slow down the computer you're running on.

Jupyter on uhura / the computing lab

On uhura, jupyter only works with chrome, so you need to start it with

jupyter notebook --browser="google-chrome"

Running jupyter remotely

Running jupyter on uhura remotely from your own laptop can be quite slow. In this case, you can speed things up significantly by running it remotely, but the browser locally (thanks to Rad for figuring this out!):

Requirements

Jupyter notebook installed in both the local_host and remote_host

Procedure

  1. Login to the remote host (e.g. uhura):
    local_host$ ssh -L localhost:8888:localhost:8889 usernm@remote_host
    If you get a message such as cannot listen to port: 8888, you can choose a different port number.
  2. cd to the Directory where you have want to create your .ipynb files and start Jupyter notebook at port 8889 on remote host without a browser:
remote_host$ jupyter notebook --no-browser --port=8889
  1. Go to a browser in the local machine and in the address bar type:
localhost:8888/
  1. If a token is requested then copy the token from the URL printed by the Jupyter notebook call in the terminal. You can also use the following command to view the tokens currently open:

In a separate session of ssh type:

remote_host$ jupyter notebook list

This should give an output like:

Currently running servers:
http://localhost:8889/?token=fccb1b2c8b7157de14466993c5d374c2933f6a4a3e284c53

Copy and paste the token. You should be able to run jupyter notebook faster and on the local browser without window forwarding.

Side effects

Once a port is assigned to remote_host it is no longer available to the local_host. So in the above example port 8888 is now binded to the remote host. This means that a local jupyter notebook cannot be started with the same port which also happens to be the default port.

Hence the following command may not work on the local machine:

local_host$ jupyter notebook

The workaround is very similar to the example above i.e. using a different port available.

local_host$ jupyter notebook port=8889

Currently, the only way I know to reset all ports is to reboot the local machine.