This workshop is intended to provide an introduction to data science tools and approaches for instructors with a background in R. The goal is to help identify key capacities for instructors and students to "think with data" while answering a statistical question. Key topics include data wrangling (operations to get data in the right form for graphics), reproducibility, web-scraping, and constructing static and dynamic graphics from data.

• Nicholas Horton, Amherst College
• Danny Kaplan, Macalester College

Supported by NSF DUE # 0920350, Project MOSAIC.

USCOTS 2015, May 27-28, 2015

• Introduce you to some key aspects of data science
  • enough for you to become a plausible, independent data scientist.
• Help you become proficient with software and techniques for carrying out data science.
• Help you, as educators to
  • use data science to enliven your statistics classes
  • make informed decisions about the extent to which data science can become part of your institution's curriculum

• The structure of data [tidy data]
• Data wrangling to condense and combine data sources to a form suitable for display
• Visualization, one technique for displaying data informatively
• "Machine learning"
• Acquiring data: scraping and cleaning
• An intro to dynamic, interactive graphics.
• Reproducible workflow for communicating, updating, and revising your work.

Each of the workshop topics is important enough that there is a variety of software implementing approaches to it.

We focus on a particular set of interrelated software tools based in R & RStudio. This software "stack" is already very widely used professionally and use is growing rapidly.

1. Structure of data: tidyr et al.
2. Wrangling: dplyr et al.
3. Visualization: ggplot2 et al.
4. Dynamics graphics: shiny
5. Reproducible workflow: knitr & rmarkdown

Much less consensus on unified tools for machine learning or for acquiring data. So we use a hodgepodge. We expect that by USCOTS 2017 these will be fully integrated into the stack.

NB: (1), (2), and (3) are sometimes called the "Hadley Stack."

Statistics

There's a lot of overlap between statistics and data science. We figure you already know a lot about statistics so we don't cover it here.

• Statistical tests, e.g., the "Stat 101 Stack"
• Statistical models, e.g. generalized linear models

But we will do a bit with machine/statistical learning

Enterprise scale data

• Database design
• Unstructured data
• Data large enough that special network techniques are needed.
• Streaming data