# -*- mode: org -*-
#+TITLE:     Jupyter : tips and tricks, installing and configuring
#+AUTHOR:    Arnaud Legrand, Benoit Rospars, Konrad Hinsen
#+DATE: June, 2018
#+STARTUP: overview indent
#+OPTIONS: num:nil toc:t
#+PROPERTY: header-args :eval never-export

* Table of Contents                                                     :TOC:
- [[#jupyter-tips-and-tricks][Jupyter tips and tricks]]
  - [[#creating-or-importing-a-notebook][Creating or importing a notebook]]
  - [[#running-r-and-python-in-the-same-notebook][Running R and Python in the same notebook]]
  - [[#other-languages][Other languages]]
- [[#installing-and-configuring-jupyter-on-your-computer][Installing and configuring Jupyter on your computer]]
  - [[#installing-jupyter][Installing Jupyter]]
  - [[#making-sure-jupyter-allows-you-to-use-r][Making sure Jupyter allows you to use R]]
  - [[#additional-tips][Additional tips]]

* Jupyter tips and tricks
The following [[https://www.dataquest.io/blog/jupyter-notebook-tips-tricks-shortcuts/][webpage]] lists several Jupyter tricks (in particular, it
illustrates many =IPython magic= commands) that should improve your
efficiency (note that this blog post is about two years old so some of
the tricks may have been integrated in the default behavior of Jupyter
now).
** Creating or importing a notebook
Using the Jupyter environment we deployed for this MOOC will allow to
easily access any file from your default GitLab project. There are
situations however where you may want to play with other notebooks.
- Adding a brand new notebook in a given directory :: Simply follow
     the following steps:
  1. From the menu: =File -> Open=. You're now in the Jupyter file manager.
  2. Navigate to the directory where you want your notebook to be created.
  3. Then from the top right button: =New -> Notebook: Python 3=.
  4. Give your notebook a name from the menu: =File -> Rename=.

     N.B.: If you create a file by doing ~File -> New Notebook ->
     Python 3~, the new notebook will be created in the current
     directory. Moving it afterward is possible but a bit cumbersome
     (you'll have to go through the Jupyter file manager by following
     the menu =File -> Open=, then select it, =Shut= it =down=, and =Move=
     and/or =Rename=).
- Importing an already existing notebook :: If your notebook is
     already in your GitLab project, then simply synchronize by using
     the =Git pull= button and use the =File -> Open= menu. Otherwise,
     imagine, you want to import the [[https://app-learninglab.inria.fr/gitlab/moocrr-session2/moocrr-reproducibility-study/blob/master/src/Python3/challenger.ipynb][following notebook]] from someone
     else's repository to re-execute it.
  1. Download the file on your computer. E.g., for this [[https://app-learninglab.inria.fr/gitlab/moocrr-session2/moocrr-reproducibility-study/blob/master/src/Python3/challenger.ipynb][GitLab hosted
     notebook]], click on =Open raw= (a small =</>= within a document icon)
     and save (=Ctrl-S= on most browsers) the content (a long JSON text
     file).
  2. Open the Jupyter file manager from the menu =File -> Open= and
     navigate to the directory where you want to upload your notebook.
  3. Then from the top right button, =Upload= the previously downloaded
     notebook and confirm the upload.
  4. Open the freshly uploaded notebook through the Jupyter file
     manager.
** Running R and Python in the same notebook
It used to be impossible with earlier versions of Jupyter but it is
now very easy thanks to the the =rpy2= package (see the details of the
installation procedurer in the corresponding section below) that
allows you to use both languages in the same notebook. Simply open a
new python notebook and follow these instructions:
1. Loading =rpy2=: 
   #+begin_src python :results output :exports both
   %load_ext rpy2.ipython
   #+end_src
2. Using the =%R= Ipython magic:
   #+begin_src python :results output :exports both
   %%R 
   summary(cars)
   #+end_src
   Python objects can then even be passed to R as follows (assuming =df=
   is a pandas dataframe):
   #+begin_src python :results output :exports both
   %%R -i df
   plot(df)
   #+end_src
Note that this =%%R= notation indicates that R should be used for the whole cell but
an other possibility is to use =%R= to have a single line of R within a
python cell.

[[https://gitlab.inria.fr/learninglab/mooc-rr/mooc-rr-ressources/blob/master/documents/notebooks/notebook_Jupyter_Python_R.ipynb][Here]] is an notebook example using both R et Python

** Other languages
Jupyter is not limited to Python and R. Many other languages are available:
[[https://github.com/jupyter/jupyter/wiki/Jupyter-kernels][https://github.com/jupyter/jupyter/wiki/Jupyter-kernels]], including
non-free languages like SAS, Mathematica, Matlab... Note that the maturity of these kernels differs widely.

None of these other languages have been deployed in the context of our
MOOC but you may want to read the next sections to learn how
to set up your own Jupyter on your computer and benefit from these extensions.

Since the question was asked several times, if you really need to stay
with SAS, you should know that SAS can be used within Jupyter using
either the [[https://sassoftware.github.io/sas_kernel/][Python SASKernel]] or the [[https://sassoftware.github.io/saspy/][Python SASPy]] package (step by step
explanations about this are given [[https://gitlab.inria.fr/learninglab/mooc-rr/mooc-rr-ressources/blob/master/documents/tuto_jupyter_windows/tuto_jupyter_windows.md][here]]). 

Since proprietary software such as SAS cannot easily be inspected, we discourage its use as it hinders reproducibility by
essence. But perfection does not exist anyway and using Jupyter
literate programming approach allied with systematic control version
and environment control will certainly help anyway.

* Installing and configuring Jupyter on your computer
In this section, we explain how to set up a Jupyter environment on
your own computer similar to the one deployed for this MOOC.

Note that Jupyter notebooks are only a small part of the picture and
that Jupyter is now part of a bigger project: [[https://blog.jupyter.org/jupyterlab-is-ready-for-users-5a6f039b8906][JupyterLab]], which allows
you to mix various components (including notebooks) in your
browser. In the context of this MOOC, our time frame was too short to
benefit from JupyterLab which was still under active development. You may, however, prefer JupyterLab when doing an installation on your own computer.

** Installing Jupyter
Follow these instructions if you wish to have a Jupyter environment on
your own computer similar to the one we set up for this MOOC.

First, download and install the [[https://conda.io/miniconda.html][latest version of Miniconda]]. We use
Miniconda version =4.5.4= and Python version =3.6= on our server.

Miniconda is a light version of Anaconda, which includes Python, the Jupyter Notebook, and other commonly used packages for scientific computing and data science.

Then download the [[https://gist.github.com/brospars/4671d9013f0d99e1c961482dab533c57][mooc_rr environment file]] and create the environment using conda:
#+begin_src shell :results output :exports both
conda env create -f environment.yml

# Windows activate the environment
activate mooc_rr

# Linux and MacOS activate the environment
source activate mooc_rr

# Linux, MacOS and Windows: launch the notebook
jupyter notebook
#+end_src
** Making sure Jupyter allows you to use R
The environment described in the last section should include R, but if
you proceeded otherwise and only have Python available in Jupyter, you
may want to read the following section.
*** • Installing [[https://github.com/IRkernel/IRkernel][IRKernel]] (R package)
IRKernel will allow you to manage notebooks using natively R rather
than python. We assume in this section that you already have a working
installation of R

*Windows*

On Windows, you are unlikely to have the tools allowing you to compile
and install the latest version of IRKernel and you should rather
install the binary version. In an R console (or in Rstudio), type the
following commands:
#+begin_src R :results output :session *R* :exports both
install.packages('IRkernel',dep=TRUE)
IRkernel::installspec()  # to register the kernel in the current R installation
#+end_src

Right below, you will find a few explanations on how to use an R
notebook. 

*Linux or Mac*

To install the latest version of IRkernel, open an R console or
Rstudio.
- Install the =devtools= package:
  #+begin_src R :results output :session *R* :exports both
  install.packages('devtools',dep=TRUE)
  #+end_src
- Define a proxy if needed (this is important if you are working in a
  company which limits the access to the Internet):
  #+begin_src R :results output :session *R* :exports both
  library(httr)
  set_config(use_proxy(url="proxy", port=80, username="username", password="password")) 
  #+end_src

- Install the =IRkernel= package:
  #+begin_src R :results output :session *R* :exports both
  devtools::install_github('IRkernel/IRkernel')
  IRkernel::installspec()  # to register the kernel in the current R installation
  #+end_src

You will then be able to create native R notebooks:
[[file:jupyter_images/new_notebook.png]]

[[file:jupyter_images/notebook_R.png]]

Note the R icon in the top right corner. [[https://gitlab.inria.fr/learninglab/mooc-rr/mooc-rr-ressources/blob/master/documents/notebooks/notebook_Jupyter_R.ipynb][Here is an example of R
notebook]].
*** • Installing rpy2 (Python package)
The =rpy2= package allows python to seamlessly call R and therefore to
have both languages in the same notebook.

*Linux or Mac*

On Linux, the rpy2 package is available in standard distributions. For
example on debian or ubuntu:
#+begin_src shell :results output :exports both
sudo apt-get install python3-rpy2 python3-tzlocal
#+end_src
An alternative (not really recommended if the first one is available)
consists in going through the python package manager with
#+begin_src python :results output :exports both
pip3 install rpy2
#+end_src

*Windows*

Download the =rpy2= [[https://www.lfd.uci.edu/~gohlke/pythonlibs/#rpy2][binary file]] by choosing the right operating system.

Open a DOS console, move to the /download/ directory and type the following command:
#+begin_src shell :results output :exports both
python -m pip install rpy2‑2.9.4‑cp37‑cp37m‑win_amd64.whl # adapt filename
#+end_src

Install also =tzlocal=:
#+begin_src shell :results output :exports both
python -m pip install tzlocal
#+end_src

If you ever run into troubles, you may want to have a look on
[[https://stackoverflow.com/questions/14882477/rpy2-install-on-windows-7][StackOverflow]] (NB : when we tried it, there has been no need to define
the =R_HOME= and =R_USER= environnement variables).

You should be able to run Jupyter and to create a python notebook that runs
R commands by following the instructions given in the beginning of
this document (look for =rpy2=).
** LaTeX for PDF export
For exporting your notebooks as PDF files, you must also install LaTeX on your system. We describe this process in a [[https://www.fun-mooc.fr/courses/course-v1:inria+41016+session02/jump_to_id/19c2b1de7766484bae73f3ab133463c6][separate resource]].
** Additional tips
*** • Exporting a notebook
Here is what we had to install on a recent Debian computer to make sure
the notebook export via LaTeX works:
#+begin_src shell :results output :exports both
sudo apt-get install texlive-xetex wkhtmltopdf
#+end_src

Obviously, you can convert to html or pdf using the using the =File > Download as > HTML= (or =PDF=) menu option.  This can also be done from
the command line with the following command:

#+begin_src sh :results output :exports both
ipython3 nbconvert --to pdf Untitled.ipynb
#+end_src

If you want to use a specific style, then the =nbconvert= exporter
should be customized. This is discussed and demoed [[http://markus-beuckelmann.de/blog/customizing-nbconvert-pdf.html][here]]. We encourage
you to simply read the [[https://nbconvert.readthedocs.io/en/latest/][doc of nbconvert]]. 

Instead of going directly through LaTeX and playing too much with the
=nbconvert= exporter, an other option consists in exporting to Markdown
and playing with [[https://pandoc.org/][pandoc]]. Both approaches work, it's rather a matter of
taste.

*Windows*

Download and install MiKTeX from the [[https://miktex.org/download][MiKTeX webpage]] by choosing the
right operating system. You will be prompted to install some specific
packages when exporting to pdf.

*** • Improving notebook readability
Here are a few extensions that can ease your life:
- [[https://stackoverflow.com/questions/33159518/collapse-cell-in-jupyter-notebook][Code folding]] to improve readability when browsing the notebook.
  #+begin_src shell :results output :exports both
  pip3 install jupyter_contrib_nbextensions
  # jupyter contrib nbextension install --user # not done yet
  #+end_src
- [[https://github.com/kirbs-/hide_code][Hiding code]] to improve readability when exporting.
  #+begin_src sh :results output :exports both
  sudo pip3 install hide_code
  sudo jupyter-nbextension  install --py hide_code
  jupyter-nbextension  enable --py hide_code
  jupyter-serverextension enable --py hide_code
  #+end_src
  
*** • Interacting with GitLab and GitHub
To ease your experience, we added pull/push buttons that allow
you to commit and sync with GitLab. This development was specific to
the MOOC but inspired from a previous [[https://github.com/Lab41/sunny-side-up][proof of concept]]. We have
recently discovered that someone else developed about at the same time
a [[https://github.com/sat28/githubcommit][rather generic version of this Jupyter plugin]]. Otherwise, remember
that it is very easy to insert a shell cell in Jupyter in which you
can easily issue git commands. This is how we work most of the time.

This being said, you may have noticed that Jupyter keeps a perfect
track of the sequence in which cells have been run by updating the
"output index". This is a very good property from the reproducibility
point of view but depending on your usage, you may find it a bit
painful when committing. Some people have thus developed [[https://gist.github.com/pbugnion/ea2797393033b54674af][specific git
hooks]] to ignore these numbers when committing Jupyter notebooks. There
is a long an interesting discussion about various options on
[[https://stackoverflow.com/questions/18734739/using-ipython-notebooks-under-version-control][StackOverflow]].

For those who use [[https://blog.jupyter.org/jupyterlab-is-ready-for-users-5a6f039b8906][JupyterLab]] rather than the plain Jupyter, a specific [[https://github.com/jupyterlab/jupyterlab-git][JupyterLab git plugin]] has been developed to offer a nice version control experience.