[R] RMarkdown vignettes v. Jupyter notebooks?

Yihui Xie x|e @end|ng |rom y|hu|@n@me
Thu Oct 11 16:15:08 CEST 2018

I just have one comment on the multi-language support in R Markdown
(inline below):

On Thu, Oct 11, 2018 at 6:19 AM Ista Zahn <istazahn using gmail.com> wrote:
> Hi Spencer,
> On Thu, Oct 11, 2018 at 5:08 AM Spencer Graves
> <spencer.graves using effectivedefense.org> wrote:
> >
> > Hello:
> >
> >
> >        What are the differences between Jupyter notebooks and RMarkdown
> > vignettes?
> Here are some of the main differences I'm aware of:
> Rmarkdown files include code and prose. The results produced by the
> code do not appear in the Rmarkdown file; instead, the file must be
> processed and typeset to produce a .pdf or .html (etc.) file that
> includes those results. Jupyter notebooks dispense with with
> processing step: output is displayed directly in the notebook. That
> is, a notebook includes code, prose, and results, while an Rmarkdown
> file includes only code and prose.
> Rmarkdown allows control over the inclusion of code and results, and
> over how the results are displayed via header arguments. There is no
> such thing in Jupyter notebooks. Controlling the precise appearance of
> a document produced by Jupyter is much more difficult. Going farther,
> one can even say that Jupyter notebooks are designed primarily to be
> read as Jupyter notebooks; exporting to other formats is kind of an
> afterthought. In contrast, Rmarkdown is designed to produce the final
> readable result in another format (.html or .pdf typically).
> Rmarkdown is based on markdown, a human readable markup language,
> Jupyter notebooks are based on JSON, a data interchange format common
> on the web. This means that Rmarkdown files can be easily edited using
> any text editor you like. The same is not true of Jupyter notebooks.
> While you can of course edit the JSON directly, the format is designed
> to be written and read by a computer; editing it yourself is not easy.
> Rmarkdown is specific to R (I guess there is some basic support in
> knitr for other languages, but in my experience it never worked well)
> while Jupyter notebooks are language agnostic and "kernels" exist for
> a large number of programming languages. However, each Jupyter
> notebook can use only one kernel; you can't easily have R and Python
> code in the same notebook.

You might want to at least update your impression about the Python
support in R Markdown now :) Last year, with the release of the
reticulate package, the support for Python has been substantially
enhanced in both R Markdown
(https://bookdown.org/yihui/rmarkdown/language-engines.html) and
RStudio (https://blog.rstudio.com/2018/10/09/rstudio-1-2-preview-reticulated-python/).
You can easily have both R and Python code in the same R Markdown
document, and the two worlds can freely talk to each other.

> Jupyter notebooks typically run in your browser where the actual text
> editing features are somewhat limited. Rmarkdown is typically run in
> an editor such as Emacs or Rstudio where editing and project support
> is much better and greater customization may be possible. You can work
> indirectly with Jupyter notebooks in Emacs
> (https://github.com/millejoh/emacs-ipython-notebook) and perhaps other
> editors as well; this goes some way toward escaping the tyranny of the
> browser but is more fragile and difficult to get working compared to
> Rmarkdown.
> Because Jupyter uses a web-based client-server model, it is easy to
> provide live interactive notebooks on your website (see e.g.,
> https://github.com/jupyterhub/binderhub). As far as I know this is not
> currently possible with Rmarkdown.
> >
> >
> >        I'm trying to do real time monitoring of the broadcast quality of
> > a radio station, and it seems to me that it may be easier to do that in
> > Python than in R.[1]  This led me to a recent post to
> > "python-list using python.org" that mentioned "Jupyter, Mathematica, and the
> > Future of the Research Paper"[2] by Paul Romer, who won the 2018 Nobel
> > Memorial Prize in Economics only a few days ago.  In brief, this article
> > suggests that Jupyter notebooks may replace publication in refereed
> > scientific journals as the primary vehicle for sharing scientific
> > research, because they make it so easy for readers to follow both the
> > scientific and computational logic and test their own modifications.
> >
> >
> >        A "Jupyter Notebook Tutorial: The Definitive Guide"[3] suggested
> > I first install Anaconda Navigator.  I got version 1.9.2 of that.  It
> > opens with options for eight different "applications" including
> > JupyterLab 0.34.9, Jupyter Notebook 5.6.0, Spyder 3.3.1 (an IDE for
> > Python), and RStudio 1.1.456.
> >
> >
> >        This leads to several questions:
> >
> >
> >              1.  In general, what experiences have people had with
> > Jupyter Notebooks, Anaconda Navigator, and RMarkdown vignettes in
> > RStudio, and the similarities and differences?  Do you know any
> > references that discuss this?
> I've used both extensively, and noted the differences I've discovered above.
> >
> >
> >              2.  More specifically, does it make sense to try to use
> > RStudio from within Anaconda Navigator, or is one better off using
> > RStudio as a separate, stand alone application -- or should one even
> > abandon RStudio and run R instead from within a Jupyter Notebook? [I'm
> > new to this topic, so it's possible that this question doesn't even make
> > sense.]
> The only advantage I can think of to using Rstudio via Anaconda is
> that you could use conda environments to maintain different versions
> or R and/or R packages for different projects.
> You'll have to weigh the pros and cons to decide whether to switch
> from Rstudio to Jupyter notebooks. Depending on what you want to do
> there are both advantages and disadvantages, as discussed above.
> Finally, I have to give a plug for a couple of related tools that I
> find very useful.
> Emacs org-mode https://orgmode.org/ gives you the best of both worlds:
> notebooks unconstrained by the browser that can include code in
> multiple languages, header arguments, excellent export support, etc.
> It is superior to both Jupyter and Rmarkdown, except that support only
> exists in Emacs.
> Jupytext (https://github.com/mwouts/jupytext) is another way to have
> it all, by allowing you to edit in markdown or Rmarkdown, and
> auto-generating a notebook and possibly other formats for you. I've
> only recently started experimenting with it, but so far I like it a
> lot.
> Best,
> Ista
> >
> >
> >        Thanks,
> >        Spencer Graves
> >
> >
> > [1] If you have ideas for how best to do real time monitoring of
> > broadcast quality of a radio station, I'd love to hear them.  I need
> > software that will do that, preferably something that's free, open
> > source.  The commercial software I've seen for this is not adequate for
> > my purposes, so I'm trying to write my own.  I have a sample script in
> > Python that will read a live stream from a radio tuner and output a
> > *.wav of whatever length I want, and I wrote Python eight years ago for
> > a similar real time application.  I'd prefer to use R, but I don't know
> > how to get started.
> >
> >
> > [2] 2018-04-13:
> > "https://paulromer.net/jupyter-mathematica-and-the-future-of-the-research-paper".
> > This further cites a similar article in The Atlantic from 2018-04-05:
> > "www.theatlantic.com/science/archive/2018/04/the-scientific-paper-is-obsolete/556676".

More information about the R-help mailing list