Functions useful for exploratory data analysis using random forests. Developed by Zachary M. Jones and Fridolin Linder in support of "Exploratory Data Analysis Using Random Forests."
This package extends the functionality of random forests fit by party (multivariate, regression, and classification), randomForestSRC (regression, classification, and survival), and randomForest (regression and classification).
Functionality includes:
partial_dependencewhich computes the expected prediction made by the random forest if it were marginalized to only depend on a subset of the features.extract_proximity(supervised) andrandomforest_distance(unsupervised) which compute the distance between observations on the training data or new data.variable_importancewhich computes feature importance for arbitrary loss functions, aggregated across the training data or for individual observations. This may also be used for subsets of the feature space in order to detect interactions.extract_proximityandplot_proxwhich computes or extracts proximity matrices and plots them using a biplot given a matrix of principal components of said matrix
Pull requests, bug reports, feature requests, etc. are welcome!