Algorithms for supervised learning of penalty functions for change detection

This R package provides a practical implementation of the algorithms described in our ICML’13 paper, Learning Sparse Penalties for Change-point Detection using Max Margin Interval Regression. This package can be used to learn a penalty function for predicting highly accurate changepoints in labeled data sets such as neuroblastoma. The main advantage of this method is increased accuracy, relative to unsupervised methods (such as information criteria: AIC, BIC, etc). In particular, since we have labels that indicate where there should and should not be changepoints, we can use them to learn a highly accurate penalty function which minimizes the number of incorrectly predicted labels. The learned penalty function can be used for predicting the number of changepoints in other data (even in data without labels).

Installation

From CRAN:

install.packages("penaltyLearning")

From GitHub:

if(!require(devtools))install.packages("devtools")
devtools::install_github("tdhock/penaltyLearning")

Usage

See my useR2017 tutorial on Supervised changepoint detection [video], which explains how to use penaltyLearning in detail. [GitHub repo with links to additional course materials]