Reproducible Analyses with

R Markdown & IPython Notebook

Wagner Lab Meeting, 10/31/14

Benefits of Reproducible Analyses

  • Keep track of analysis steps going from the raw data to final statistics & plots
  • Update stats & plots based on revisions to analysis (or addition of subjects)
  • Makes sharing analysis pipelines easier
    • Collaboration
    • Publication
    • Replication
  • Avoid errors in working between multiple programs & copy/paste mistakes

How?

How?

  • R Markdown
  • IPython Notebook
  • MATLAB Publishing

MATLAB Publishing

  • Compile HTML, PDF, etc. documents from an R script
  • Improve readability of code by formatting text using Markdown syntax
  • Create interactive R Markdown documents using Shiny

Why use it?

How?

open source!

MATLAB Publishing

Quick notes

# Generate HTML view of script
publish('run_analysis.m');

# View HTML
web('html/run_analysis.html')
  • File Exchange scripts for including dependent and called functions in published version
  • Con: not integrated with statistical analyses in R

R Markdown

  • Compile HTML, PDF, etc. documents from an R script
  • Improve readability of code by formatting text using Markdown syntax
  • Create interactive R Markdown documents using Shiny

Why use it?

R Markdown

  • Download/Install R
  • Download/Install RStudio (optional, but recommended)
  • Install the "knitr" package

How do you get it?

http://stanford.edu/class/psych252/tutorials/index.html

R Markdown

  • Headers
  • Emphasis (*italics*, **bold**)
  • LaTeX equations ($\chi^2$)
  • Tables
  • Lists
  • Images (!(/path/to/image.png)
  • Code chunks/code blocks
  • Inline code (``)

Markdown Basics

R Markdown

  • To execute a line of code, type: cmd-return
  • To execute a chunk: opt-cmd-c
  • To insert a new chunk: opt-cmd-i
  • For code completion: tab
  • For documentation: ?func-name; ??keyword
  • Comment/uncomment line: ​shift-cmd-c

Keyboard Shortcuts

R Markdown

R Code Chunks

```{r load_data}
behav_filepath = '~/Dropbox/Code/tutorial/objfam_groupcat_euc.csv'

# Load data
df = read.csv(behav_filepath)

# Examine data structure
summary(df)
str(df)
```

R Markdown

R Code Chunks 

```{r load_data}
behav_filepath = '~/Dropbox/Code/tutorial/objfam_groupcat_euc.csv'

# Load data
df = read.csv(behav_filepath)

# Examine data structure
summary(df)
str(df)
```

Naming chunks helps navigate script

R Markdown

R Code Chunks

R Markdown

R Code Chunks 

Plot data
----------
- Specify height & width of plot
- Hide code
```{r plot_data, fig.height=3, fig.width=7, echo=FALSE}
boxplot(Response~Morph, data=df)
```

Specify plot dimensions

hide code in knit file

R Markdown

R Code Chunks 

General linear model analysis
----------------------
### Does Euclidean distance vary as a function of morph level?
```{r stats}
rs1 = lm(EuclidDist~scale(Morph, scale=FALSE), data=df)
summary(rs1)

final_model = rs1
sm = summary(final_model)
```

#### Stats Summary:
Euclidean distance significantly varies as a function of morph level, 
$R^2$ = `r sm$r.squared`, 
*F*(`r sm$fstatistic['numdf']`, `r sm$fstatistic['dendf']`) = 
`r sm$fstatistic['value']`, 
*t* = `r sm$coefficients[6]`, 
*estimate* = `r sm$coefficients[2]`.

LaTeX style text & inline code

R Markdown

Interactive Documents

```{r, echo = FALSE}
sliderInput("morph", "Morph level:", min = 1, max = 3, value = 2)

renderPlot({
  # Subset data
  df_subset = df[df$Morph==input$morph,]
  
  # draw the histogram with the specified morph level
  with(df_subset, hist(Response, 
                       col = 'darkgray', 
                       border = 'white'))
})
```

input to plot

R Markdown

  • Github (.Rmd)
    • Collaboration with the .Rmd file
    • Version control

How to share it?

R Markdown

  • Github (.Rmd)
    • Collaboration with the .Rmd file
    • Version control
  • RPubs (html)
    • Prettier formatting

How to share it?

R Markdown

  • Github (.Rmd)
    • Collaboration with the .Rmd file
    • Version control
  • RPubs (html)
    • Prettier formatting

How to share it?

Hands-on Rmd tutorial:

https://github.com/sgagnon/kemeny/blob/master/R/wagner-lab-meeting.Rmd​

https://rpubs.com/sgagnon/wagner-lab-meeting​

R Tutorials

  • Basic R data structures (vectors, lists, dataframes)
  • Manipulating dataframes
    • integer/numeric vs. factor
    • reshaping (long vs. short)
    • accessing rows/columns
  • Plotting
    • base graphics
    • ggplot2
http://stanford.edu/class/psych252/tutorials/index.html

Basics

R Tutorials

http://stanford.edu/class/psych252/plots/index.html

http://www.cookbook-r.com/

On the web

http://tryr.codeschool.com

http://www.statmethods.net

https://www.datacamp.com/courses/introduction-to-r

http://swirlstats.com

Thorough Tutorials:

Quick Resources:

https://twitter.com/rstudiotips

Other:

http://adv-r.had.co.nz/

R Tutorials

  • Coursera Course: The Data Scientist's Toolbox
    • Nov 3 - Dec 1 (3 hrs/week)
    • Practical introduction to the tools like version control, markdown, git, GitHub, R, and RStudio.

On the web

R Tutorials

IPython Notebook

  • Interactive environment where you can combine code, output, plots, images, markdown, in one document
  • Allows use of all Python packages, including Pandas, scipy, numpy, nibabel, etc.
  • Integrated with R using rpy2
  • Easy plotting with matplotlib & Michael's seaborn package
  • Easy to share documents with nbviewer

Why use it?

IPython Notebook

  • Download/Install Anaconda

  • Install packages via pip

 

  • Make sure the environment variable PATH is set to the Anaconda location

How do you get it? 

export PATH=~/anaconda/bin:$PATH
pip install <package-name>

IPython Notebook

  • Open up a Terminal window, navigate to the location where you'd like the notebook stored, and type:
  • The homepage for your current directory will open in the browser; click on a current notebook to open, or select New Notebook.

How do you use it?

ipython notebook

IPython

Examples

  • Each block can be code (default), a heading, or markdown
  • To execute code, click on the block and type: shift-return
  • For code completion: tab; for tooltips inside a function: shift-tab
  • For more information: ??, e.g., sns.factorplot??

IPython

Examples

  • Command mode (esc)
    • convert to markdown: m
    • or to heading: 1,2,3...6
    • select previous/next cell: up/down arrows
    • new cell above/below: a/b
    • delete cell: dd
  • Edit mode (return)

IPython

Examples

  • Interactive widgets
    • Allows for interactive exploration of data within the notebook

IPython

Examples

    • R in Python​
      • ​rpy2
        • IPython ​magic integration
pip install rpy2
# Load R extension
%load_ext rpy2.ipython

# Import dataframe df to R
%R -i df

# New code block:
%%R
print(str(df))

IPython

Examples 

pip install rpy2

IPython

Useful Packages

  • Pandas
    • Open source library for data structures & data analysis
  • Numpy
    • Efficient multi-D arrays
    • Scientific computing tools
  • Matplotlib
    • 2D plotting library, MATLAB-like interface
  • Seaborn (Michael)
    • ​Statistical data visualization based on matplotlib

IPython

Pandas

  • DataFrame object for easy data manipulation
  • Flexible reshaping/pivoting of datasets
  • Intelligent label-based slicing, subsetting of large datasets
  • Merging/joining of multiple datasets
  • Hierarchical indexing
  • Tools for reading & writing data between different formats (CSV, text, excel, SQL)

Hands-on IPyNB tutorial:

NBViewer: html

Github: .ipynb file

Copy of reproducible-analyses

By Steph Gagnon

Made with Slides.com

Copy of reproducible-analyses

Reproducible analyses with IPython and RMarkdown

  • 957

More from Steph Gagnon