This repository contains code for the useR2017 tutorial “Introduction to optimal changepoint detection algorithms” by Toby Dylan Hocking and Rebecca Killick.
Rebecca: Unsupervised changepoint detection 90 minutes video. Source Rmd.
Toby: Supervised changepoint detection NAU copy, 90 minutes video. Source Rmd.
- Do the exercises during class on this Rcloud notebook https://rcloud.social/edit.html?notebook=3e66297efb401bc0c6333ad517804f85 (make sure to click the fork so you can makes changes). source: Supervised-exercises.R
For a more advanced tutorial, check out coarseDataTools::EMforCFR, adapEnetClass, icenReg, and non-linear models.
Try `cv.glmnet(family=”poisson”)` for predicting the number of changepoints.
Copy and adapt tutorial materials from Rebecca’s eRum2016 workshop.
depmixS4.models.R tries to fit an HMM to the neuroblastoma data set, but I ran into depmixS4.bugs.R – I emailed the author <[email protected]> on May 31 but I haven’t heard anything yet.
Compare with BIC from PELT/fpop pelt.fpop.R
Compiled Rmd HTML Supervised changepoint tutorial, first draft, source: Supervised.Rmd.
montreal-biohackathon-2017.org describes a changepoint detection challenge for a 24 hour Biohackathon.
figure-regression-interactive-some.R creates 5 plot interactive data viz.
figure-regression-interactive-some.R for interactive figure with a few profiles that we can zoom in to. http://bl.ocks.org/tdhock/raw/9fc37a7aaf291cef364aab3fb41dd898/
figure-regression-interactive.R for comparing BIC and learned model on entire neuroblastoma data set.
penaltyLearning package for exactModelSelection and targetInterval.
Begin breakpoint.learning.cv.R which will read breakpoint.learning.RData and estimate breakpoint predicted test error via 6-fold cross-validation (including BIC, mBIC, supervised penalty learning via iregnet).