Conditional Visualisation for
Statistical Models
This is the website for the condvis package in R. This package produces interactive graphics for visualising 2-D and 3-D sections in data space, showing both fitted models and observed data.
You can install the current CRAN release of condvis with
install.packages("condvis")condvis is for making low-dimensional visualisations of statistical models in high-dimensional space.
Taking sections is a sensible way to display and understand objects embedded in high-dimensional space. The condvis package allows us to directly visualise a complex regression surface (e.g., a random forest or gradient boosted tree model) in high-dimensional data space. This is achieved by taking sections (slices) through the space, or put another way, by holding all but one or two model inputs constant and allowing the remaining input(s) to vary across their range.
Often, we like to see our fitted model and observed data on the same plot, to see if the model is doing the job we want it to do. When continuous predictors are involved, sections in data space very rarely contain much, if any, observed data points. A natural thing to do is to display nearby data points.
The condvis package is free, open source software licenced under GPL-2 or later. The R logo is used on this website under CC-BY-SA 4.0.