This course focuses on the concepts and tools behind reporting modern data analyses in a reproducible manner. Reproducible research is the idea that data analyses, and more generally, scientific claims, are published with their data and software code so that others may verify the findings and build upon them. The need for reproducibility is increasing dramatically as data analyses become more complex, involving larger datasets and more sophisticated computations. Reproducibility allows for people to focus on the actual content of a data analysis, rather than on superficial details reported in a written summary. In addition, reproducibility makes an analysis more useful to others because the data and code that actually conducted the analysis are available. This course will focus on literate statistical analysis tools which allow one to publish data analyses in a single document that allows others to easily execute the same analysis to obtain the same results.
knitr (part 1)
Loading...
Из курса от партнера Johns Hopkins University
Reproducible Research
2665 оценки
Из урока
Week 2: Markdown & knitr
This week we cover some of the core tools for developing reproducible documents. We cover the literate programming tool knitr and show how to integrate it with Markdown to publish reproducible web documents. We also introduce the first peer assessment which will require you to write up a reproducible data analysis using knitr.
Познакомьтесь с преподавателями
Roger D. Peng, PhDAssociate Professor, Biostatistics
Bloomberg School of Public Health
Jeff Leek, PhDAssociate Professor, Biostatistics
Bloomberg School of Public Health
Brian Caffo, PhDProfessor, Biostatistics
Bloomberg School of Public Health
This lecture's going to be about literate statistical programming.
In the lecture about reproducible research, I talked about the basic
ideas of what it means for an analysis to be reproducible.
But now I'm going to talk about one of the tools that you can use to actually make
your analysis reproducible by others, or at least
help other peoples to kind of reproduce your analysis.
And one of the ways that we talked about
making reproducible documents was to put this literate statistical
programming, literate statistical programming
concept, or just literate programming.
And so the basic idea is that, you know, authors have to
undertake a lot of effort to put data and results on the web.
And even when they put that stuff on the web, they, they're all kind of separate.
The data's over here, the code is over here, and you kind of
have to figure out well, which code goes with whi, which data
and how do you, wha, how do you escape this code and
do you load this first, do you do this first, and so,
there can be a lot of confusion even when everything is just available.
And, and so.
One of the ways you can simply that process is to
put the data, and the code together in the same document
so to speak so that people can execute the code in
the right order, and the data are read at the right times.
So you can have a single document that integrates the data analysis with all
the textural representations so the words that you use to describe was going on.
And so you don't have that everything kind
of separate pieces but everything's kind of linked together.
So, the, the, the original idea of how literate program comes from Don Knuth.
He was a computer scientist at Stanford and
his original system was for writing just regular
computer programs and, and documenting the computer code
at the same time as writing the code.
So the basic idea translates over to a literate statistical
program where you want to document your analysis and similarly
have the code for your analysis in the same document.
And so the ideas and article's a string of text and
code and analysis code is divided into text and code chunks.
And so the presentation code formats the tables, formats figures, things like that.
There's article text that talks about what's going on.
And then these literate programs, there's two concepts, you can weave
them into readable, human readable documents, and you can tangle them to
produce machine readable documents or just code files.
So, literate programming, as I said before, is a general concept.
You need a documentation language and a programming language.
So the original Sweave system that I talked about before, written by Fritz
Leisch, used LaTeX as the documentation
language and R as the programming language.
So on this lecture I'm going to talk
about knitr, which supports a variety of documentation and
programming languages.
And so the first thing you want to start with, so
the basic question is, how do I make my work reproducible?
And the basic answer to that is that you have to decide to do it, okay.
And you can decide to do this at any point in your, in
your analysis, but it's usually easiest to do it at the very beginning.
If you decide at the very end that you want to make your
analysis reproducible, then it often is much
harder to do and maybe even impossible.
And, and so we want to keep track of things as you go along.
You can use a version control system like Get of SVN.
I won't talk about that here but if you've
heard about that that is a good thing to learn.
It's important to use software, statistical
software, whose operation can be coded.
So the idea is that you can write down the
instructions that were used to manipulate or analyze the data.
So this generally rules
out graphical user interfaces unless those programs will code or keep track
of all the clicks that you make on the graphical user interface.
Systems like R and other types of
packages, because you have to program them explicitly.
The code, as long as you save the code, will be, will kind of,
will have everything that you, will store everything that you did to the data.
Another key lesson is to not save any output.
And by output,
I mean mostly kind of temporary data transformation types of [UNKNOWN].
So if you pre-process the data, and have a kind of clean data set.
Rather than store the clean data set, it might be better just
to have the kind of raw data set with the preprocessing code.
because then you, you not only do you have, can you
make the final product, but you can see how you got there.
If you just keep the clean data set, and, and then you accidentally
lose the preprocessing code, then you don't really have a good record of
how you got from A to B.
And so try to keep, rather than keeping the temporary products or even final
products, try to keep the original products and the code that got you there.
And that will make it easier to understand what you did
and how you got to the data that you ended up analyzing.
So not saving up is I think a key element of reproducibility.
And try not to save data in non-proprietary formats.
These are data formats that are not where the, where the lay
out of the data set is not publicly known.
There are not too many proprietary formats out there any
more that I think are commonly used like they were before.
But a lot of them, for example, products will
store data and because it's just much more efficient.
In a proprietary format, but then that makes it difficult to transfer
to another person if that person doesn't have the same exact program.
Then they won't be able to read the data.
And so using non-proprietary,
even textual formats may be compressed can be much better at, at making,
much much, can make your re, research a little bit, or analysis more reproducible.
than, proprietary formats.
So there are a couple of pros and cons when
it comes to the literate programming style, this is not
the only talk you can use to make your work
reproducible, but literate programming is just one of the tools.
And so, some of the pros of that; it forces
you to put all the text and the code of your
analysis in one place, so it occurs in a logical order.
All the data and the results are automatically updated to
reflect external changes, so if you, kind of, so you have
this live document, and you make some changes, when you reprocess
the document, it will automatically update itself to reflect any changes.
So you don't have, you don't have a situation where
numbers will be out of date or things will be mismatch,
mismatched.
Because the code will kind of automatically recalculate everything.
Because the code is live in the document, a necessity you
have to run the code in order to produce the final document.
That's the weeding process.
You get this kind of regression test so to speak, not like linear regression, but
regression test to see so if there are
any mistakes that you introduced into the analysis.
If the code doesn't run then you know you've made, you've
made an error, you've introduced a problem so the, so the running
of the live code will kind of keep
you from introducing new errors into the analysis.
Some of the cons when it comes to literate programming
are that of course the text and the code are
all in one place so in particular if you have
a lot of code or a very complex lengthy analysis.
Then you're like a human readable text is going to be mixed in with all this
code and it can make it difficult to
edit the document because you have to somehow search
through all the, this whole document to figure
out where the text is amongst all the code.
So that can be a problem for the author.
Also, if your analysis is very complicated and you have to re-process
this document every time you see a human readable version of it.
It can, it can a bit substantially slow
down the process by which you build the document.
Ознакомьтесь с нашим каталогом
Присоединяйтесь бесплатно и получайте персонализированные рекомендации, обновления и предложения.
Coursera делает лучшее в мире образование доступным каждому, предлагая онлайн-курсы от ведущих университетов и организаций.
© Coursera Inc., 2018 Все права защищены.
