Philip Data Management notes

Philip's notes on DasGupta webinar, 14 Sept 2015:

webfeet is Dasgupta's gitbub name.  presentation source will go on github
  
His template for folders
  lib: background stuff
    R: custom functions, one per file
    src: C C++ code
    tests: code to test functions
    packages.R: list of necessary libraries
    reload.R: what is necessary to reload all data sets
    scripts: custom scripts for Python file manipulation and graphing
  toplevel stuff:
    DataAcquisition.r: producing data sets
    DataMunging.r: cleaning data
    Figures.r: All graphs
    Modeling.r: all model fitting
    Report.Rmd: R Markdown to produce report
    
His philosophy:
  "Do the common things once" (and only once)
  So put one function definition (or very closely related functions) per file
    Copy the file to new project when you need that function again
    
Documentation   
  write comments to help you search for code you need 
  
Software carpentry: Greg Willson (sp?)
  disciplined approach/philosophy for code management

Jeff Leek's datasharing advice on GitHub: written for his students

Hadley's tidy data paper in J Statistical Software
  tidyverse: collection of packages for tidy data
  grope: Robinson package to convert results into tidy data sets
  
Coding style suggestions: 
  Google has internal R and Python document
  Hadley has R document
  both on the web
  
Testbadge? in R: package for unit testing of functions

Rashomon effect: each model tells you something different, truth is the overall gestalt
  (PMD: sounds like informal model averaging)
  
broom and estout: R packages to write results to excel tables

Jupyter notebooks: now include R. "can do some pretty amazing things with it"

Frank Harrell has Sweave package for clinical trial reporting

Keynote makes html presentations
  
ReporteRs translating a table into a Word table

Karen Healy (blog) how to use literate programming to collaborate with others