class: center, middle, inverse, title-slide # <font class = "title-panel"> PPOL564 | Data Science I - Foundations </font> <font size=100, face="bold"> Reproducibility</font> <br> <br> ### <font class = "title-footer"> Prof. Eric Dunford ◆ Georgetown University ◆ McCourt School of Public Policy ◆ <a href="mailto:eric.dunford@georgetown.edu" class="email">eric.dunford@georgetown.edu</a></font> --- layout: true <div class="slide-footer"><span> PPOL564 | Data Science I - Foundations           Week 2 <!-- Week of the Footer Here -->              Work Flow and Reproducibility<!-- Title of the lecture here --> </span></div> --- ## We focus on things like this... <!-- <img src="figures/lecture01_ethnic-exposure-boko-haram.png" width="600"> --> .center[] --- ## And forget the reality that is this... .center[<img src="figures/lecture01_messy-files.png" width="800">] --- ## Reproducibility with a captial "R" **...is fundamental to the scientific method, but it is also a <u> practical reality </u>.** <br> - juggling multiple versions of the same file - collaboration can create conflicts across versions - projects are picked up and put down → tracing the progression of a project across a spiderweb of files is not always easy (or possible) - new people enter the fray → getting them up-to-speed means walking them through the labyrinth, which wastes time and resources. --- # Generating Reproducible Work <br> ### 1. Readable ### 2. Portable ### 3. Well-Named ### 4. Repeatable ### 5. Version Control --- ## Readable ```python x = np.random.normal(size=100) y = 1 + 2*x + np.random.normal(size=100) plt.scatter(x,y) ``` vs. ```r # Monte Carlo Simulation of a bivariate linear regression sample_size = 100 # simulated sample size indep_var = np.random.normal(size=sample_size) # independent variable error = np.random.normal(size=sample_size) # simulate error # generate dependent variable as function of the # independent variable and some error. dep_var = 1 + 2*indep_var + error # plot values plt.scatter(indep_var,dep_var) ``` ??? - Well Commented Code and Functions** - Well-Named Objects - Leverage Spacing To a degree, Code (like writing) should be more Hemmingway than Faulkner: concise, clear, readable. --- ## Portable - **Project can easily travel across computers** - Python's Virtual Environments (`venv`) and R Project (`.rproj`) - **Scripts avoid "machine" specific designations** + Avoid **specific file paths**: `/Users/my-user-name/data-projects/my-project` + **Retain software and packages versions**. - **Use text files** + Not software dependent (e.g. .docx, .ia); Can open on any system + Can be easily searched via the commandline + Easy to track changes via version control --- # Well-named - **No spaces!** + A space between designations can mean many things + spaces are ambiguous for the computer + `data analysis 2.py` → `data-analysis-2.py` - **Names that state the purpose of the file** (no matter how long). + `data-analysis-2.py` → `Analysis01_wrangling-census-data-for-visualization_v2.py` --- # Well-named - Maintain **designated folders** for different aspects of the project. ```bash data-project ├── raw-data/ # Where our input data lives ├── output-data/ # Where our manipulated data lives ├── py/ # Where our Python functions live ├── R/ # Where our R functions live ├── figures/ # Where our generated figures live ├── reports/ # Where our text-based (.tex/.md/.txt) live └── analysis/ # Where our analyses live ``` --- ## Repeatable <br><br> - **Every step of the project can be expressed as code** - **Automate what you can** - **Use functions to repeat common tasks** - **Clearly state all dependencies** (i.e. packages/modules) at the top of every script --- ## Version Control <br> - **Retain a record of all changes** made throughout the **_project's lifespan_** - **Easily handle collaboration**: + track who did what + uniform method dealing with conflicting changes - **Provides a room for experimentation and non-linear exploration** - No more **version file names**!