stepR (version 2.1-10)

stepcand: Forward selection of candidate jumps

Description

Find candidates for jumps in serial data by forward selection.

Usage

stepcand(y, x = 1:length(y), x0 = 2 * x[1] - x[2], max.cand = NULL,
  family = c("gauss", "gaussvar", "poisson", "binomial", "gaussKern"), param = NULL,
  weights = rep(1, length(y)), cand.radius = 0)

Value

An object of class stepcand extending class stepfit such that it can be used as an input to steppath.stepcand: additionally contains columns

cumSum

The cumulative sum of x up to rightEnd.

cumSumSq

The cumulative sum of squares of x up to rightEnd (for family = "gauss").

cumSumWe

The cumulative sum of weights up to rightEnd.

improve

The improvement this jump brought about when it was selected.

Arguments

y

a numeric vector containing the serial data

x

a numeric vector of the same length as y containing the corresponding sample points

x0

a single numeric giving the last unobserved sample point directly before sampling started

max.cand

single integer giving the maximal number of blocks to find; defaults to using all data (note: there will be one block more than the number of jumps

family

distribution of the errors, either "gauss", "poisson" or "binomial"; "gaussInhibit" is like "gauss" forbids jumps getting close together or to the ends in steppath.stepcand, "gaussInhibitBoth" already forbids this in stepcand (not recommended)

param

additional parameters specifying the distribution of the errors; the number of trials for family "binomial"; for gaussInhibit and gaussInhibitBoth a numeric of length 3 with components "start", "middle" and "end" preventing the first jump from getting closer to x0 than the "start" value, any two jumps from getting closer than the "middle" value, and the last jump from geting closer than the "end" value to the end, all distances measured by weights (cf. example below)

weights

a numeric vector of the same length as y containing non-negative weights

cand.radius

a non-negative integer: adds for each candidate found all indices that are at most cand.radius away

See Also

steppath, stepfit, family

Examples

# simulate 5 blocks (4 jumps) within a total of 100 data points
b <- c(sort(sample(1:99, 4)), 100)
f <- rep(rnorm(5, 0, 4), c(b[1], diff(b)))
rbind(b = b, f = unique(f), lambda = exp(unique(f) / 10) * 20)
# add gaussian noise
x <- f + rnorm(100)
# find 10 candidate jumps
stepcand(x, max.cand = 10)
# for poisson observations
y <- rpois(100, exp(f / 10) * 20)
# find 10 candidate jumps
stepcand(y, max.cand = 10, family = "poisson")
# for binomial observations
size <- 10
z <- rbinom(100, size, pnorm(f / 10))
# find 10 candidate jumps
stepcand(z, max.cand = 10, family = "binomial", param = size)