Conda for teaching

You are going to give a teaching course, and you need a specific code environment.

Setup

First - find a good location that everyone can read (and not write). I'd suggest somewhere under /cm/shared/apps/SHARED/ as a starting point - this allows everyone to access this location. It's important not to put anything secret there - it's a public resource, so please bear that in mind.

Next - create a folder for your environment:

mkdir /cm/shared/apps/SHARED/my_conda_env
chmod +r /cm/shared/apps/SHARED/my_conda_env

You may want to manipulate the permissions for this folder if someone is going to set this up with you. Consider the commands in Shared folders.

Then, install Anaconda into it (choose the year and month yourself!):

Also, note that you need to make the sh script executable before you can run it.

wget https://repo.anaconda.com/archive/Anaconda3-YEAR.MONTH-Linux-x86_64.sh
./Anaconda3-YEAR.MONTH-Linux-x86_64.sh -s -b -p /cm/shared/apps/SHARED/my_conda_env 

Now you have a working conda environment in this folder. You can manipulate this here by running /cm/shared/apps/SHARED/my_conda_env/bin/conda, or, I would recommend creating a modulefile so that you can use it as default.

Create the following example modulefile in a matching /cm/shared/modulefiles/SHARED/my_conda_env:

#%Module -*- tcl -*-
##
## conda environment modulefile
##

set                     loadedmodules           [split $::env(LOADEDMODULES) ":"]
set                     modulepath              [split $ModulesCurrentModulefile "/"]
set                     envpath                 [lrange $modulepath 4 end]

set                     root                    /cm/shared/apps/[join $envpath "/"]

proc ModulesHelp { } {
        global version

        puts stderr "\tThis module provides the conda environment at $envpath"
}


if { [module-info mode] != "whatis" } {
        puts stderr "[module-info mode] environent $envpath ."
}

module-whatis   "Provides environment $envpath"


prepend-path            PATH                    $root/bin
prepend-path            LD_LIBRARY_PATH         $root/lib
prepend-path            LIBRARY_PATH            $root/lib
prepend-path            CPATH                   $root/include
prepend-path            MANPATH                 $root/share/man

This will allow you to module load SHARED/my_conda_env and thus have conda pathed to the currently active environment.

Jupyter Kernel

In order for students to be able to use this environment in jupyter, they will need a kernel definition.

Kernel definitions are usually a separate folder containing, in particular, a file called kernel.json, plus an icon that is displayed that represents this kernel, and other helper code.

The setup for this is that you should create the following folder for their access:

mkdir -p /cm/shared/apps/SHARED/my_conda_env/kernel/my_conda_env
chmod +r /cm/shared/apps/SHARED/my_conda_env/kernel/my_conda_env

This folder is something they will need to copy in place to their home directory, specifically $HOME/.local/share/jupyter/kernels/

Inside this folder, create the following kernel.json file. Watch out that the paths will need to match the current environment path if you're using a different location!

Python Kernel

vim /cm/shared/apps/SHARED/my_conda_env/kernel/my_conda_env/kernel.json
</source>
<source lang='bash'>
{
 "env": {
   "PATH":

"/cm/shared/apps/SHARED/my_conda_env/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin"
},
 "language": "python",
 "argv": [
  "/cm/shared/apps/SHARED/my_conda_env/bin/python",
  "-m",
  "ipykernel",
  "-f",
  "{connection_file}"
 ],
 "display_name": "my_conda_env"
}

For Python, that's it. Ipykernel is installed automatically on conda initialisation.

R Kernel

For an R kernel, you need to make sure that the IRkernel package is installed. This is the package that is used to communicate from Jupyter to your running R kernel.

#MISSING EXAMPLE CODE
#PROBABLY /cm/shared/apps/SHARED/my_conda_env/bin/conda install R_irkernel

You'll also need to create two files: kernel.json and kernel.js. the kernel.js is a helper script to allow jupyter to communicate to R effectively:

vim /cm/shared/apps/SHARED/my_conda_env/kernel/my_conda_env/kernel.json

{
 "env": {
   "LD_LIBRARY_PATH":
      "/cm/shared/apps/SHARED/my_conda_env/lib:/cm/shared/apps/SHARED/my_conda_env/lib64"
 },
  "argv": ["/cm/shared/apps/SHARED/my_conda_env/bin/R", "--slave", "-e", "IRkernel::main()", "--args", "{connection_file}"],
  "display_name": "MAE50806-AdvMolEcol/Sandbox_R",
  "language": "R"
}

vim /cm/shared/apps/SHARED/my_conda_env/kernel/my_conda_env/kernel.js

const cmd_key = /Mac/.test(navigator.platform) ? 'Cmd' : 'Ctrl'

const edit_actions = [
	{
		name: 'R Assign',
		shortcut: 'Alt--',
		icon: 'fa-long-arrow-left',
		help: 'R: Inserts the left-assign operator (<-)',
		handler(cm) {
			cm.replaceSelection(' <- ')
		},
	},
	{
		name: 'R Pipe',
		shortcut: `Shift-${cmd_key}-M`,
		icon: 'fa-angle-right',
		help: 'R: Inserts the magrittr pipe operator (%>%)',
		handler(cm) {
			cm.replaceSelection(' %>% ')
		},
	},
	{
		name: 'R Help',
		shortcut: 'F1',
		icon: 'fa-book',
		help: 'R: Shows the manpage for the item under the cursor',
		handler(cm, cell) {
			const {anchor, head} = cm.findWordAt(cm.getCursor())
			const word = cm.getRange(anchor, head)
			
			const callbacks = cell.get_callbacks()
			const options = {silent: false, store_history: false, stop_on_error: true}
			cell.last_msg_id = cell.notebook.kernel.execute(`help(\`${word}\`)`, callbacks, options)
		},
	},
]

const prefix = 'irkernel'

function add_edit_shortcut(notebook, actions, keyboard_manager, edit_action) {
	const {name, shortcut, icon, help, handler} = edit_action
	
	const action = {
		icon, help,
		help_index : 'zz',
		handler: () => {
			const cell = notebook.get_selected_cell()
			handler(cell.code_mirror, cell)
		},
	}
	
	const full_name = actions.register(action, name, prefix)
	
	Jupyter.keyboard_manager.edit_shortcuts.add_shortcut(shortcut, full_name)
}

function render_math(pager, html) {
	if (!html) return
	const $container = pager.pager_element.find('#pager-container')
	$container.find('p[style="text-align: center;"]').map((i, e) =>
		e.outerHTML = `\\[${e.querySelector('i').innerHTML}\\]`)
	$container.find('i').map((i, e) =>
		e.outerHTML = `\\(${e.innerHTML}\\)`)
	MathJax.Hub.Queue(['Typeset', MathJax.Hub, $container[0]])
}

define(['base/js/namespace'], ({
	notebook,
	actions,
	keyboard_manager,
	pager,
}) => ({
	onload() {
		edit_actions.forEach(a => add_edit_shortcut(notebook, actions, keyboard_manager, a))
		
		pager.events.on('open_with_text.Pager', (event, {data: {'text/html': html}}) =>
			render_math(pager, html))
	},
}))

Last Steps

In order to help your students get their kernel definitions into $HOME/.local/share/jupyter/kernels/, it's probably a good idea to write a small and simple notebook to do this when executed, or else instruct them to do:

cp -rv /cm/shared/apps/SHARED/my_conda_env/kernel/* .local/share/jupyter/kernels/