kshedden / formula

Formula processing in Go

formula : Formula parser for Go

formula is a library for building data sets in Go using formulas. The most common use-case for this library is to build design matrices for use in statistical regression analysis. The formulas in this package behave much like formulas in R, Julia, Matlab, and Python (using Patsy). Interactions, algebraic expansion, and dummy-coding are all supported. Compared to these other formula packages, there are a few simplifying differences:

• Only one-sided formulas are supported. Multiple formulas can be parsed together to produce a single dataset. To produce a dataset for a regression model, parse two formulas at once -- one formula for each side of the regression relationship.

• Main effects are not automatically included for interactions. Include them manually as desired.

• Functions (transformations) must be deterministic, not "stateful"

Design: The data to be processed using formulas must be accesed through a DataSource, which is a simple interface that allows slices to be retrieved by name. Parsing one or more formulas produces a ColSet, which contains all the variables resulting from parsing the formula(s). A ColSet is interchangeable with a statmodel.Dataset, so can be passed directly into that package for modeling.

Modeling: This package does not fit any statistical models. If you want to fit a model to the dataset produced by the formula package, you can use one of these packages: GLM, duration.

See here for examples that use this package to produce datasets, and that then use the statmodel package to fit regression models.