Kenny Flagg

Data Scientist : Statistician : Dancer : Aspiring Nomad : Amateur Musician

I put in a lot of effort to learn the tools I use, and I want to make it easier for the next generation.

Python

While working in business data science, I migrated from R to Python. I began an ongoing side project (likely never to be finished) to create a curriculum for people with statistical training to learn to do data analysis in Python. Certain sections are complete enough to be useful.

  • If your exposure to “programming” is instructors providing you with code that they themselves barely understand, Chapter 1 will be valuable.
  • If you have done some coding in Python and need experience working with tabular data, Chapter 2 curates a set of Pandas tutorials.
  • Similarly, Chapter 3 will get you starting doing some classical statistics.

For data visualization, I prefer Plotnine (a Python/Pandas port of GGPlot2). Seaborn is another popular module which produces nice output. Matplotlib is ubiquitous, but I have extremely little patience for it.

Now, if you are a budding data scientist, you will at some point need to ensure that the code you wrote gets run the exact same way by the exact same version of every module on every single machine that ever runs it. I don’t want to get that deep or technical here, but this article on Pyenv is something I have consulted many times.

R

There are many modern resources for learning R properly. Wickham et al’s R for Data Science is thorough and a classic.

For some of the things that I compiled for my own work:

  • Tutorial on fitting spatial LGCP models with R-INLA and a finite element (SPDE) approximation.
  • PoisDPmix, an R package for fitting infinite mixtures of Gaussians to partially-observed spatial point pattern data.
    • The code and documentation are in a poor state as I do not expect this to be useful to anyone, but if you want to use it please let me know!