The document discusses the requirements for supporting Jupyter notebooks for proposed geolocation activities in a course. It notes that the current notebooks require internet access to call third party APIs and use a non-standard Python package. This adds to the requirements of a basic Jupyter notebook distribution. Options discussed for meeting these requirements include customizing the Python kernel, having students install packages, or defining a Python distribution with the additional package pre-installed.

1. Requirements associated with suppor3ng the use of Jupyter notebooks for proposed
geoloca3on ac3vi3es in TM112.
1

2. Jupyter notebook are a widely used technology for documen3ng computa3onal
scripts…
2

3. …and sharing interac3ve, computa3onal, educa3onal resources.
3

4. Current draD of geoloca3on ac3vi3es: hFps://github.com/psychemedia/OU-TM11N/
blob/master/d2/geocodingDemos.ipynb
Whilst the notebokk describes a simple set of related ac3vi3es, they require that the
notebook server has access to the wider internet and is capable of making calls to
third party APIs over hFp.
It also makes use of a python package that is not part of the standard Jupyter
notebook/scien3fic python distribu3on and must be installed either by the student or
provided as part of an enhanced python distribu3on accessed via the notebook
server.
The ac3vity defini3on thus adds two addi3onal requirements to the provision of a
basic, “self-contained” Jupyter notebook distribu3on designed to make use of a
standard scien3fic python distribu3on:
1) That the kernel has access to the wider internet in order to access third party APIs
(and, as a side effect, external python module repositories such as pip).
2) Either that the student user has permissions to install an addi3onal python
module if the required addi3onal package is not installed, or that the course
team can deine a python distribu3on with the addi3onal required package
4

5. Pushing at the boundary condi3ons IS DELIBERATE. In part, solving the problems
introduced herein means that Jupyter notebooks can be used in a flexible, open
manner to support courses in the future, and provide students with an opportunity
to interact with third party services as well as closed world, OU based ac3vi3es.
5

6. The notebooks provide the ac3vity descrip3ons and are used by a notebook server to
provide the interface through which the ac3vi3es can be completed in an interac3ve
manner.
hFps://github.com/psychemedia/OU-TM11N/blob/master/d2/
geocodingDemos.ipynb
6

7. hFps://github.com/psychemedia/OU-TM11N/blob/master/d2/
geocodingDemos.ipynb
The notebook needs to be able to access third party APIs.
Note that this limit also applies to accessing datafiles by URL, which means data files
cannot be loaded directly into a notebook kernel from a URL, but must be
downloaded onto the user’s desktop and uploaded by hand into the notebook
environment if they are to be processed using the notebook.
7

8. The notebook needs access to a non-standard python package. Either this needs to
be installed in a customised python kernel, or students need permission to install the
required package.
The notebook requires network access to fetch map 3les used to render an
embedded map.
8

9. hFps://github.com/psychemedia/OU-TM11N/blob/master/d2/
geocodingDemos.ipynb
One of the ac3vi3es allows students to look up the loca3on of a network router from
its MAC address. If the student is running the notebook kernel locally, and their
computer has wifi enabled, they should be able to iden3fy a local wifi router MAC
address of their own and use that as part of the exercise.
If the kernel is running on a remote server, it is unlikely that wifi networking services
will be available, in which case we can provide students with an alterna3ve MAC
address to look up.
9

10. 10

11. The TM351 Jupyter notebook server includes several customisa3ons, including:
- OU branding of notebooks;
- Custom exports: MicrosoD Word .docx, ODSzip (a zip file containing the original
notebook and the HTML rendering of it)
11

12. 12

13. A Jupyter notebook can be viewed in a browser as either a sta3c document, or as an
interac3ve user interface to a code execu3ng kernel running as a process on the local
machine, or on a remote host.
Embedded code outputs can include responses from third party APIs, data tables,
charts, interac3ve maps generated from executed code.
13

14. The notebook source file (suffix .ipynb) is a JSON text file.
It can be rendered to an HTML document using the `nbviewer` applica3on.
The notebook file can also be used interac3vely, as a GUI to a backend computa3onal
process that can execute and evaluate elements iden3fied as “code” in the notebook,
and return the results of code execu3on for display in the notebook.
--
{
ipynb[label=".ipynbn(JSON text file)"];
nbviewer[label="nbviewer",color='yellow'];
ipynb -> nbviewer;
group {
shape=line;
html1[label="HTML",color='lightgreen'];
nbviewer -> html1;
14

15. The notebook server can be run variously:
- on the student’s desktop as a desktop applica3on, e.g. using Anaconda (for
example, in FutureLearn Learn to Code MOOC);
- in a virtual machine running on the student’s desktop eg as in the TM351
Virtualbox VM;
- on a remote host, such as in a Docker container running on Azure;
- on a remote host, launched via the Jupyterhub mul3-user notebook server
manager. In this case, the notebook may launch as a “disposable” notebook server
(for example, using tmpnb or try.jupyter.org) or may require an authen3cated
login.
15

16. What sort of issues are relevant to understanding our an3cipated student experience
of working with the notebooks?
16

17. How do I “run” an interac3ve notebook?
- If you are running a local notebook server, you need to install it on your desktop
somehow, have some means for star3ng and stopping the server, and know what
local URL to find the notebooks on. (In some circumstances the notebook server
may automa3cally open the notebook server homepage in the user’s browser).
- If you are accessing a remote server, you need to know where to find the
notebook server and may need authen3ca3on creden3als.
Do I need to run the notebook server on my computer?
- The wifi lookup ac3vity will require the notebook server to run on a student’s
computer that has support for wifi connec3vity. All other ac3vi3es could run on a
remote server, as long as the python kernel has access to the wider internet.
Do I need a network connec3on?
- To access the notebooks running on a remote server, you will need a network
connec3on in order to work with the notebooks interac3vely, At the end of a
session, you can export an HTML, docx or PDF version of the notebook to keep a
“flat” (non-interac3ve) porwolio record of it.
17

18. There are several ways students *could* run Jupyter notebooks – so let’s quickly
review them… Some approaches may work for some course situa3ons, some may
not…
Along the way of reviewing the various offerings, can we come up with some sort of
criteria for helping us iden3fy what requirements different sorts of ac3vi3es might
impose, and what sorts of requirements can be met by the different ways of
accessing the notebooks.
18

19. There are several online services that provide access to online Jupyter notebooks.
19

20. tmpnb.org / try.jupyter.org
The temporary notebook server launches a temporary container pre-seeded with
demo notebooks.
The backend servers are provided (free to the end-user) by Rackspace.
Users can upload their own notebooks and execute them in a container launched for
that individual.
The containers are automa3cally destroyed, along with any content uploaded to
them or created by the notebook, aDer a short period of 3me.
20

21. tmpnb.org / try.jupyter.org
Users cannot install addi3onal packages or call out to the internet to access third
party APIs or URLs.
21

22. notebooks.azure.com
MicrosoD are aggressively trying to grow an audience for their online data science
tools, and notebooks.azure.com are a complement to that service, offering free
Jupyter notebooks in a project based environment.
Notebooks can be collected together in libraries and shared with third par3es,
offering the possibility of placing all the notebooks required for a par3cular course
into one project library and allows students to clone that library in order to access
the notebooks.
Notebooks are saved to the file area between sessions.
22

23. notebooks.azure.com
Access to the internet from the notebook kernel appears to be off-limits, meaning no
access to third party APIs or URLs.
23

24. CoCalc is a rebranded version of SageMathCloud.
24

25. CoCalc does not provide access to the wider internet on the free plan, but an paid for
upgrade does allow calls to be made to external services [UNTESTED].
We have used SageMathCLoud / CoCalc as one of the ways of accessing Jupyter
notebooks in the FutureLearn course Learn to Code for Data Analysis.
25

26. Commercial upgrades are available at personal and course level.
26

27. There are several ways of running Jupyter notebooks on your own computer.
27

28. Anaconda - hFps://www.con3nuum.io/downloads - is a scien3fic Python distribu3on
that bundles Jupyter notebooks, which can be started from the command line or a
shortcut launcher.
Students can download and install the distribu3on on their own computer, and then
install the addi3onal required package.
Anaconda was the primary means by which we encouraged learners to access Jupyter
notebooks on the FutureLearn Learn to Code for Data Analysis course.
28

29. hFps://github.com/nteract/nteract
nteract is a cross-plaworm HTML5 (electron) desktop applica3on that runs a
notebook server ad provides access to notebooks in a standalone applica3on. Double
clicking on a ipynb file will open it in nteract. Although a standalone applica3on,
nteract depends on the presence of an appropriate IPython kernel, such as one
installed with Anaconda.
nteract is s3ll under development and has not had an official release to date.
29

30. hFps://www.virtualbox.org/wiki/Downloads
VirtualBox is a cross-plaworm virtualisa3on applica3on that can be used to run
predefined virtual machines (VMs).
A VirtualBox headless VM can be defined to run the Jupyter notebook server as a
service, accessed via the browser on the student’s own desktop. Notebooks exist in a
folder on the host machine and shared into the VM.
The VM approach was used by TM351 to deliver a notebook server along with several
other applica3ons to students. A USB installer included all set-up files and the VM
image, although it was also possible to download this from original loca3ons on the
web.
30

31. Docker community edi3on allows you to run docker containers on your own desktop.
Docker now runs as a na3ve applica3on on Windows and Macs, handling container
virtualisa3on without the need to install VirtualBox.
31

32. I’m thinking here – fire up your own host and use OU provided docker container
images, Amazon AMI images etc? eg hFps://github.com/psychemedia/ou-tm351/
blob/master/tm351_builder_unversioned/docs/tm351_ami.md
32

33. Launch your own instance but at the generosity of a third party.
Eg mybinder
The mybinder server tends to be unreliable – but could the OU host it’s own, that
only launches whitelisted images from authen3cated OU student users?
33

34. Eg run notebook container on OU OpenStack; advantages: notebook can be seeded
with branded notebook server, export extensions, course notebooks
34

35. Eg run notebook container on OU OpenStack; advantages: notebook can be seeded
with branded notebook server, export extensions, course notebooks
35

36. Provide a temporary notebook server running somewhere, access limited to TM112
students, pre-seeded course notebooks, Jupyter notebook environment customised
with OU branding, export extensions, required python packages preinstalled.
Netwrok access controls from original tmpnb server would need to be relaxed if calls
were to be made to third party APIs.
Alterna3vely, OU could also host API endpoint servers to handle geocoding, map
3leserver, and notebooks could be configured to access those OU provided services.
36

37. Hosted on OU iron or Azure?
Students have authen3cated access to their own notebook server via Jupyterhub.
This will support individual persistence of files.
37

38. 38

39. Provide a mac address to try exercise with
39

40. Maps won’t render, calls to third party APIs will fail.
Weak alterna3ve is to make executed notebook available so students can read
through outputs generated as canned/example outputs by us. This could even be as
float HTML, docx, PDF etc.
Alterna3ve is to provide linked services running local API endpoints – eg there is
postocodes.io container, there is a 3leserver available that runs with OpenStreetMap
3les. These could be provided:
- Within a monolithic Virtualbox VM (cf. TM351 VM)
- As a set of linked docker-compose containers.
VM/container could be downloaded from internet or shipped via USB memory s3ck.
40

41. Weak alterna3ve is to make executed notebook available so students can read
through outputs generated as canned/example outputs by us. This could even be as
float HTML, docx, PDF etc.
41

42. Addi3onal requirements for TM351:
1) Persistence of student notebooks somewhere
2) Ability to connect to linked services (postgres, mongo)
3) Ability to persist link database state
42

43. The TM351 virtual machine extends the requirements over an above a simple Jupyter
notebook server.
In par3cular, the TM351 VM bundles:
- Customisa3ons to the Jupyter notebook; preinstalled python packages; addi3onal
Linux packages.
- Addi3onal linked services that are only exposed inside the VM (PostgreSQL,
MongoDB)
- An addi3onal service with browser UIs exposed outside the VM (OpenRefine)
43

44. Iden3fying the best route for delivering services with browser UIs has wider
consequences than just making Jupyter notebooks available to TM112 students, as
we have seen in the case of TM351. So what is the wider context?
44