A quadratic time dynamic programming algorithm can be used to compute an approximate solution to the problem of finding the most likely changepoints with respect to the Poisson likelihood, subject to a constraint on the number of segments, and the changes which must alternate: up, down, up, down, etc. For more info read < http://proceedings.mlr.press/v37/hocking15.html> "PeakSeg: constrained optimal segmentation and supervised penalty learning for peak detection in count data" by TD Hocking et al, proceedings of ICML2015.
depend on penaltyLearning, delete related code in this pkg.
Suggest ggplot2 >= 2.0, new geom_tallrect implementation.
cDPA C code now computes total Poisson loss which is consistent with the PoissonLoss R function, and with other packages (Segmentor3IsBack, coseg). It has been significantly cleaned up (duplication removed).
move joint segmentation code to PeakSegJoint package.
multiSampleSegZoom C function.
R implementation of multiSampleSegSome.
Scripts up to Step4 under exec/
Error checking in binSum.
chrom and sample ratio features in exec/Step1.R
multiSampleSeg Optimal and Heuristic C code.
clusterPeaks C code.
LinDynProg.c renamed to cDPA to be consistent with paper.
Do not use GSL headers for positive infinity cost; instead use INFINITY defined in "math.h"
fista.R interval regression code.
binSum C code sets count to -1 for profiles that are too short for the number of bins requested.
binSum C code for quickly computing sums over bins of constant size.
warning when user requests X peaks but that model is infeasible.