Upgrade to Pro — share decks privately, control downloads, hide ads and more …

New tools and workflows for data analysis

New tools and workflows for data analysis

Talk at Workshop on Visualization for Big Data: Strategies and Principles, Fields Institute
https://github.com/jennybc/2015-02-23_bryan-fields-talk
http://www.fields.utoronto.ca/programs/scientific/14-15/bigdata/visualization/
Data science training, R Markdown, GitHub

Jennifer (Jenny) Bryan

February 23, 2015
Tweet

More Decks by Jennifer (Jenny) Bryan

Other Decks in Science

Transcript

  1. New tools and workflows for data analysis Dr. Jennifer (Jenny)

    Bryan Dept. of Statistics & Michael Smith Laboratories, UBC Workshop on Visualization for Big Data @ Fields Institute [email protected] @JennyBryan http://stat545-ubc.github.io @STAT545 http://www.stat.ubc.ca/~jenny/ https://github.com/jennybc
  2. The Big Data Brain Drain: Why Science is in Trouble

    http://jakevdp.github.io/blog/2013/10/26/big-data-brain-drain/ in a wide array of academic fields, the ability to effectively process data is superseding other more classical modes of research
  3. Fact: I don’t work for these companies. I don’t represent

    them. I am not an author of these packages.
  4. Paul Murrell Martyn Plummer Brian Ripley Deepayan Sarkar Duncan Temple

    Lang Luke Tierney Simon Urbanek Douglas Bates John Chambers Peter Dalgaard Seth Falcon Robert Gentleman Kurt Hornik Ross Ihaka Michael Lawrence Friedrich Leisch Uwe Ligges Thomas Lumley Martin Maechler Martin Morgan Duncan Murdoch
  5. M unge Visualise M odel Communicate Tidy Question Collect W

    ednesday, October 30, 13 Slides from Hadley Wickham's talk in the Simply Statistics Unconference http://t.co/D931Og8mq3 We can’t focus just on this!
  6. nytimes 2014-08-18 Data scientists spend 50 - 80% of their

    time mired in this more mundane labor of collecting and preparing unruly digital data, before it can be explored for useful nuggets. ... what data scientists call “data wrangling,” “data munging” ...
  7. “data science is ‘just’ statistics” “data wrangling is not statistics”

    if you value self-consistency, you can hold at most one of these opinions
  8. We cannot expect anyone to know anything we didn't teach

    them ourselves. Sarah Bryce To a very great degree, daily work by other people sounds easy -- certainly easier that what we have to do. Gretchen Rubin
  9. permission requirement to invest time in setting up tools and

    to develop proficiency “simple” descriptive stats exploration through visualization tame data from “the wild” alpha to omega: raw data to a web page/app readiness for open science and automation STAT 545 now
  10. how to organize your work? how to make work more

    pleasant for you? how to make it navigable by others? how to reduce tedium and manual processes? how to reduce friction for collaboration? how to reduce friction for communication? specific tools and habits can build alot of this into the normal coding and analysis process
  11. R ≠ RStudio RStudio mediates your interaction with R; it

    would replace Emacs + ESS or Tinn-R, but not R itself Rstudio is a product of -- actually, more a driver of -- the emergence of R Markdown, knitr, R + Git(Hub)
  12. Title (header 1, actually) ===================================== This is a Markdown document.

    ## Medium header (header 2, actually) It's easy to do *italics* or __make things bold__. > All models are wrong, but some are useful. An approximate answer to the right problem is worth a good deal more than an exact answer to an approximate problem. Absolute certainty is a privilege of uneducated minds-and fanatics. It is, for scientific folk, an unattainable ideal. What you do every day matters more than what you do once in a while. We cannot expect anyone to know anything we didn't teach them ourselves. Enthusiasm is a form of social courage. Code block below. Just affects formatting here but we'll get to R Markdown for the real fun soon! ``` x <- 3 * 4 ``` I can haz equations. Inline equations, such as ... the average is computed as $\frac{1}{n} \sum_{i=1} ^{n} x_{i}$. Or display equations like this: $$ \begin{equation*} |x|= \begin{cases} x & \text{if $x≥0$,} \\\\ -x &\text{if $x\le 0$.} \end{cases} \end{equation*} $$ <!DOCTYPE html> <html> <head> <meta http-equiv="Content-Type" content="text/html; charset=utf-8"/> <title>Title (header 1, actually)</title> <!-- MathJax scripts --> <script type="text/javascript" src="https:// c328740.ssl.cf1.rackcdn.com/mathjax/2.0-latest/ MathJax.js?config=TeX-AMS-MML_HTMLorMML"> </script> <style type="text/css"> body { font-family: Helvetica, arial, sans-serif; font-size: 14px; ... <body> <h1>Title (header 1, actually)</h1> <p>This is a Markdown document.</p> <h2>Medium header (header 2, actually)</h2> <p>It&#39;s easy to do <em>italics</em> or <strong>make things bold</strong>.</p> <blockquote> <p>All models are wrong, but some are... <p>Code block below. Just affects formatting here but we&#39;ll get to R Markdown for the real fun soon!</p> <pre><code>x &lt;- 3 * 4 </code></pre> Markdown HTML Fess up: How many of you still hand-code HTML?
  13. Title (header 1, actually) ===================================== This is a Markdown document.

    ## Medium header (header 2, actually) It's easy to do *italics* or __make things bold__. > All models are wrong, but some are useful. An approximate answer to the right problem is worth a good deal more than an exact answer to an approximate problem. Absolute certainty is a privilege of uneducated minds-and fanatics. It is, for scientific folk, an unattainable ideal. What you do every day matters more than what you do once in a while. We cannot expect anyone to know anything we didn't teach them ourselves. Enthusiasm is a form of social courage. Code block below. Just affects formatting here but we'll get to R Markdown for the real fun soon! ``` x <- 3 * 4 ``` I can haz equations. Inline equations, such as ... the average is computed as $\frac{1}{n} \sum_{i=1} ^{n} x_{i}$. Or display equations like this: $$ \begin{equation*} |x|= \begin{cases} x & \text{if $x≥0$,} \\\\ -x &\text{if $x\le 0$.} \end{cases} \end{equation*} $$ Markdown HTML
  14. How is the math getting typeset? Answer: Mathjax How painful

    is that to use? Not at all. Automagic with knitr and RStudio.
  15. What happens to equations if the reader is not connected

    to the internet? The LaTeX is displayed. No great harm.
  16. If I use Markdown, am I restricted to HTML output?

    No. pandoc = “swiss-army knife” of document conversion (RStudio will gladly install and invoke for you.)
  17. If you have an annoying process for authoring for the

    web .... or If you avoid authoring for the web, because you’re not sure how ... start writing in Markdown.
  18. R Markdown rocks ===================================== This is an R Markdown document.

    ```{r} x <- rnorm(1000) head(x) ``` See how the R code gets executed and a representation thereof appears in the document? `knitr` gives you control over how to represent all conceivable types of output. In case you care, then average of the `r length(x)` random normal variates we just generated is `r round(mean(x), 3)`. Those numbers are NOT hard-wired but are computed on-the- fly. As is this figure. No more copy-paste ... copy- paste ... oops forgot to copy-paste. ```{r} plot(density(x)) ``` Note that all the previously demonstrated math typesetting still works. You don't have to choose between having math cred and being web-friendly! Inline equations, such as ... the average is computed as $\frac{1}{n} \sum_{i=1}^{n} x_{i}$. Or display equations like this: $$ \begin{equation*} |x|= \begin{cases} x & \text{if $x≥0$,} \\\\ -x &\text{if $x\le 0$.} \end{cases} \end{equation*} $$ R Markdown rocks ===================================== This is an R Markdown document. ```r x <- rnorm(1000) head(x) ``` ``` ## [1] -1.3007 0.7715 0.5585 -1.2854 1.1973 2.4157 ``` See how the R code gets executed and a representation thereof appears in the document? `knitr` gives you control over how to represent all conceivable types of output. In case you care, then average of the 1000 random normal variates we just generated is -0.081. Those numbers are NOT hard- wired but are computed on-the-fly. As is this figure. No more copy-paste ... copy-paste ... oops forgot to copy-paste. ```r plot(density(x)) ``` ![plot of chunk unnamed-chunk-2](figure/unnamed- chunk-2.png) ... R Markdown Markdown
  19. R Markdown rocks ===================================== This is an R Markdown document.

    ```r x <- rnorm(1000) head(x) ``` ``` ## [1] -1.3007 0.7715 0.5585 -1.2854 1.1973 2.4157 ``` See how the R code gets executed and a representation thereof appears in the document? `knitr` gives you control over how to represent all conceivable types of output. In case you care, then average of the 1000 random normal variates we just generated is -0.081. Those numbers are NOT hard- wired but are computed on-the-fly. As is this figure. No more copy-paste ... copy-paste ... oops forgot to copy-paste. ```r plot(density(x)) ``` ![plot of chunk unnamed-chunk-2](figure/unnamed- chunk-2.png) ... Markdown HTML
  20. Markdown HTML foo.md foo.html easy to write (and read!) easy

    to publish easy to read in browser R Markdown foo.rmd
  21. How do to actually convert Markdown to HTML? knitr, rmarkdown

    add-on packages provide user-friendly functions RStudio makes them available via button
  22. R Markdown rocks ===================================== This is an R Markdown document.

    ```{r} x <- rnorm(1000) head(x) ``` See how the R code gets executed and a representation thereof appears in the document? `knitr` gives you control over how to represent all conceivable types of output. In case you care, then average of the `r length(x)` random normal variates we just generated is `r round(mean(x), 3)`. Those numbers are NOT hard-wired but are computed on-the- fly. As is this figure. No more copy-paste ... copy- paste ... oops forgot to copy-paste. ```{r} plot(density(x)) ``` Note that all the previously demonstrated math typesetting still works. You don't have to choose between having math cred and being web-friendly! Inline equations, such as ... the average is computed as $\frac{1}{n} \sum_{i=1}^{n} x_{i}$. Or display equations like this: $$ \begin{equation*} |x|= \begin{cases} x & \text{if $x≥0$,} \\\\ -x &\text{if $x\le 0$.} \end{cases} \end{equation*} $$ R Markdown HTML How to achieve at the command line: > library("rmarkdown") > render("foo.Rmd")
  23. R Markdown rocks ===================================== This is an R Markdown document.

    ```{r} x <- rnorm(1000) head(x) ``` See how the R code gets executed and a representation thereof appears in the document? `knitr` gives you control over how to represent all conceivable types of output. In case you care, then average of the `r length(x)` random normal variates we just generated is `r round(mean(x), 3)`. Those numbers are NOT hard-wired but are computed on-the- fly. As is this figure. No more copy-paste ... copy- paste ... oops forgot to copy-paste. ```{r} plot(density(x)) ``` Note that all the previously demonstrated math typesetting still works. You don't have to choose between having math cred and being web-friendly! Inline equations, such as ... the average is computed as $\frac{1}{n} \sum_{i=1}^{n} x_{i}$. Or display equations like this: $$ \begin{equation*} |x|= \begin{cases} x & \text{if $x≥0$,} \\\\ -x &\text{if $x\le 0$.} \end{cases} \end{equation*} $$ R Markdown HTML Click here.
  24. Do I have to do everything in R markdown? What

    about plain R scripts? Use rmarkdown::render() or Rstudio’s Compile Notebook button to get a satisfying stand- alone webpage based on an R script.
  25. R Markdown: Integrating A Reproducible Analysis Tool into Introductory Statistics

    Technology Innovations in Statistics Education, 8(1) Baumer, Ben, Smith College Cetinkaya-Rundel, Mine, Duke University Bray, Andrew, Smith College Loi, Linda, Smith College Horton, Nicholas J., Amherst College Publication Date: 2014 Permalink: https://escholarship.org/uc/item/90b2f5xh
  26. How do I show the world all these awesome dynamic

    HTML reports I’m creating? Easiest: Rpubs Or do whatever you usually do to get HTML on the web. Or use GitHub ....
  27. Big picture, so far: web-friendly is good various hosting platforms

    make it easy to share web- ready products with minimal effort embedding analysis and logic in source document for a report is good - huge win for reproducibility - also excellent for communication and documentation (R) Markdown + knitr (+ RStudio) make it very easy to author dynamic reports that are ready for the web
  28. disclaimer: knitr is not limited to executing R code knitr

    is not limited to processing R Markdown I just chose to focus on R and R Markdown Read more in the book or on the web: Dynamic documents with R and knitr by Yihui Xie, part of the CRC Press / Chapman & Hall R Series (2013). ISBN: 9781482203530. http://rmarkdown.rstudio.com
  29. OK you’ve got a collection of ... R scripts R

    package R Markdown files input data intermediate results figures output tables compiled reports all evolving over time how do you keep track of this?
  30. how do I put my stuff on the web? for

    the world or select collaborators? Advice to preserve sanity: Stop doing this via email, attachments, and tracking changes in Word. Get that stuff into plain text, put it under version control and get it out on the web.
  31. Version control systems (VCS) were created to help groups of

    people develop software Git, in particular, is being “repurposed” for activities other than pure software development ... like the messy hybrid of writing, coding and data wrangling
  32. “Git, provides a lightweight yet robust framework that is ideal

    for managing the full suite of research outputs such as datasets, statistical code, figures, lab notes, and manuscripts.” “... this tool can be leveraged to make science more reproducible and transparent, foster new collaborations, and support novel uses.” Ram Source Code for Biology and Medicine 2013, 8:7 http://www.scfbm.org/content/8/1/7 BRIEF REPORTS Open Access Git can facilitate greater reproducibility and increased transparency in science Karthik Ram Abstract Background: Reproducibility is the hallmark of good science. Maintaining a high degree of transparency in scientific reporting is essential not just for gaining trust and credibility within the scientific community but also for facilitating the development of new ideas. Sharing data and computer code associated with publications is becoming increasingly common, motivated partly in response to data deposition requirements from journals and mandates from funders. Despite this increase in transparency, it is still difficult to reproduce or build upon the findings of most scientific publications without access to a more complete workflow. Findings: Version control systems (VCS), which have long been used to maintain code repositories in the software industry, are now finding new applications in science. One such open source VCS, Git, provides a lightweight yet GitHub repository for this paper: https://github.com/karthik/smb_git Ram: Git can facilitate greater reproducibility and increased transparency in science. Source Code for Biology and Medicine 2013 8:7. doi:10.1186/1751-0473-8-7
  33. collaboration = the “killer app” of version control Learning Git

    has been -- and continues to be -- painful. But not nearly as crazy-making as the alternatives: - documents as email attachments - uncertainty about which version is “master” - am I working with the most recent data? - archaelogical “digs” on old email threads - uncertainty about how/if certain changes have been made or issues solved - hair-raising ZIP archives containing file salad
  34. Git repository = a bunch of files you want to

    manage in a sane way repo = repository you can set up repo ... then start your work or you can make a set of existing files and make them into a repo
  35. Git server GitHub browser- based UI Git Git client Using

    a Git client to commit (a local operation) repo repo
  36. Git server GitHub browser- based UI Git Git client Using

    Git at the command line to commit (a local operation) repo repo
  37. Git server GitHub browser- based UI Git Git client Using

    a Git client to push (local 㱺 remote) repo repo
  38. Git server GitHub browser- based UI Git Git client Using

    Git at the command line to push (local 㱺 remote) repo repo
  39. Git server GitHub browser- based UI Git Git client Operating

    on a Git repo via GitHub in the browser repo repo
  40. Git server GitHub browser- based UI Git Git client Using

    a Git client to pull (remote 㱺 local) repo repo
  41. Git server GitHub browser- based UI Git Git client Using

    Git at the command line to pull (remote 㱺 local) repo repo
  42. Many R packages are developed in the open on GitHub

    Nice option when someone tells you to “read the source”! Many R packages are developed in the open on GitHub Nice option when someone tells you to “read the source”!
  43. Many government agencies, media outlets, academic labs, etc. put their

    stuff on GitHub https://github.com/WhiteHouse https://github.com/chicago https://github.com/fivethirtyeight https://github.com/TheUpshot https://github.com/propublica/ http://ncip.github.io (NCI’s informatics program) https://github.com/LSST (Large Synoptic Survey Telescope) https://github.com/ctb (Titus Brown lab) https://github.com/lh3 (Heng Li lab)
  44. STAT 545 is an Organization on GitHub all course materials

    are posted there (public repo) all course development was done there (private repo for instructors only) each student had his/her own repo for coursework (visible only within the Organization) rough notes on set-up
  45. Comma (.csv) and tab (.tsv) delimited files are automatically rendered

    nicely in GitHub repositories Example: some Lord of the Rings data
  46. Note the contributions to STAT 545 materials from one prof,

    3 TAs, and one kind soul from the internet
  47. Files in a Git repo, even one hosted on GitHub,

    still reside on your computer Browse and edit them all you want Git has commands for communicating with the remote repository, e.g. the GitHub repo (push, pull, fetch, clone) I highly recommend using a Git GUI on your computer for making commits, syncing with the remote, etc. Reconciling and merging changes when two people make conflicting commits is not fun, but better than the alternatives
  48. Big picture, second half: sane file and project management is

    good that’s what version control does distributed file management is good excellent for 2+ people collaborating ability to browse something on the web is unreasonably powerful Git + GitHub provide a compelling solution for collaborative file wrangling; (R) Markdown and RStudio play well with Git(Hub)
  49. R markdown Git(Hub) Data wrangling, cleaning, munging Visualization (R chops,

    in general) 8 weeks 4 weeks Automation & pipelines R packages Shiny Web APIs and scraping STAT 545 = 1 semester, 3 contact hours/wk
  50. Bottom line: do something deliberate that has a good hassle:

    result ratio for you. Be open to upgrading your approach as time goes on. Keep your eyes and ears open re: new developments.