In rsangole/tidyrabbit: Tidy up Your Carret Work.

There are many different modeling functions in R. Some have different syntax for model training and/or prediction. The package started off as a way to provide a uniform interface the functions themselves, as well as a way to standardize common tasks (such parameter tuning and variable importance). --- from the caret book

The R-ecosystem is rich, anybody can bundle their knowledge into a package tat can be used by anyone. The rich documetnaiotn system can be used to describe the exact usecases, and how to use the functions. If you want to use someones package you can see how and on what way. Individual idiosyncrasies are not a real problem if you use only one package.

However for a datascientist this delvers soem problems, we are often more interested in accuracy and performance. For that usecase we apply multiple transformations on a datasets and we try several machine learning models, and then we select one or several of those for further work.

In that case we need to learn all of varying ways packages ingest data and specify models. There must be a better way.

The caret package

The Caret ( _C_lassification _A_nd _RE_gression _T_raining) package by Max Kuhn is a way to unify all those pakages. This package presents a unified interface for all packages. You can easily swap out a decision tree for a logistic regression or other model. It unifies those packages while making use of their authors hard work.

The caret abstraction layer creates (just as other R packages) objects for transformations and predictions. These artefacts are very useful if you want to apply the same action on new data. However, if you create multiple models it starts to get quite hard to keep track of all the artefacts. Transformations of data, predict objects, etcetera. One reason is my bad habbit of giving meaningless names to objects, and second because the number of objects keep growing.

Data science projects, as other projects, are often exploratory and iterative. You try and repeat, while often forgetting to delete the useless artefacts.

Tidyverse

Many people in the R world, including many novices, are now relying on the tidyverse. The tidyverse is an opinionated collection of R packages designed for data science. All packages share an underlying philosophy and common APIs ^[https://www.tidyverse.org/].

The packages in the tidyverse have functions that take a data.frame and return a data.frame. Combined with the pipe operator, originally from the magrittr package, it is now possible to write very clear code that reads in the same order as it is executed and allows you to apply multiple steps without the need to save intermediate results. I cannot stress enough how much this changes the way we do data analysis with these tools.

Data.frames

The main object for many tidyverse packages is the data.frame. A data.frame is a rectangular object where, according to tidy principles, every row is used for an observation and every columns reflects a measurement. But most importantly, you keep the things that are related together. Every column in one data.frame has the same length and every item in one column needs to be of the same type (numeric, character, factor, or list). Although not used that much a list column is incredibly handy. A list column itself does not care what you put into it, as long as the total length of the list is the same as the other columns in a data.frame. Anything that is not a character, factor, or numeric can thus be placed inside a list.

Anything. Including other data.frames.

Tidy rabbit

The tidy rabbit package takes many of the artefacts that are created by the caret package and eats them, putting these artefacts inside data frames. That way we keep the related artefacts together.