SegCost {directlabels}R Documentation

Cost of segmentation models

Description

20 segmentation models were fit to 2 simulated signals, and several different error measures were used to quantify the model fit.

Usage

data(SegCost)

Format

A data frame with 560 observations on the following 5 variables.

bases.per.probe

a factor with levels 374 7: the sampling density of the signal.

segments

numeric: the model complexity measured using number of segments.

cost

numeric: the cost value.

type

a factor with levels Signal Breakpoint Complete Incomplete Positive: how to judge model fit? Signal: log mean squared error between latent signal and estimated signal. Breakpoint: exact breakpoint error. Complete: annotation error with a complete set of annotations. Incomplete: annotation error with only half of those annotations. Positive: no negative annotations.

error

a factor with levels E FP FN I: what kind of error? FP = False Positive, FN = False Negative, I = Imprecision, E = Error (sum of the other terms).

Source

PhD thesis of Toby Dylan Hocking, chapter Optimal penalties for breakpoint detection using segmentation model selection.


[Package directlabels version 2021.1.13 Index]