Preparing Data for Supervised Classification

Nina Zumel has been polishing up new vtreat for Python documentation and tutorials. They are coming out so good that I find to be fair to the R community I must start to back-port this new documentation to vtreat for R.

vtreat is a package for systematically preparing data for supervised machine learning tasks such as classification or regression. vtreat designs a data transform that takes in messy data (with missing values, and high cardinality categorical variables) and delivers transformed data that is purely numeric and with no missing values (essentially the data format needed by most scikit-learn machine learning procedures). The transformation is designed to try and retain almost all of the information relating the explanatory variables to the dependent variable in a model usable format. This transformation can be saved and then applied to future test or application data.

If you aren’t using something like vtreat in your data science projects: you are really missing out (and making more work for yourself).

Of course all of this is easier to evaluate with examples. And that is what Nina Zumel has been working on (in addition to supervising the semantics and theory; she invented many of the techniques, so we look to her for supervision).

Our first new Python example is here: vtreat for Classification in Python.

As I said, this example came out so well I have ported it from Python to R here: vtreat for Classification in R.

If I get some free time I will also back-port vtreat for regression in Python and vtreat for unsupervised tasks in Python to R. I also would like to note an upcoming treat for R users: chapter 8 “Advanced Data Preparation” of the second edition of Practical Data Science with R (Zumel, Mount; 2019) is all about vtreat!

Edit/follow-up: I have ported more of the documentation, here is an index.