bimodalSep: A function that, given a numeric vector, finds the value...

In baySeq: Empirical Bayesian analysis of patterns of differential expression in count data

Description

This function takes a numeric vector and finds the value which splits the data into two sets of minimal total variance, weighted by the size of subsets (Otsu's method). It is principally intended to be a quick and easy way of separating bimodally distributed data.

Usage

bimodalSeparator(x, weights = NULL, minperc = 0.1, elbow = NULL)

Arguments

x	A numeric vector containing the data to be split.
weights	Possible weightings on the values in x for calculating the variance.
minperc	The required minimum size of each of the two subsets, expressed as a percentage of the total size. See Details.
elbow	If set, finds the 'left' or 'right' elbow of variance, instead of the minimum; defaults to NULL. See Details.

Details

This function is intended to give a quick and easy way of splitting bimodally distributed data. Where there are large outliers in the data, it may be that the value which minimises the variance does not split the bimodal data but isolates the outliers. The 'minperc' parameter can be used to ensure that each subset of the split data will be of some minimum size, avoiding the outlier problem.

If 'elbow = NULL' (the default) then the split occurs at the value that minimises the variance, x0. If 'elbow = left' then we attempt to find the elbow point to the left of the value that minimises the variance, if 'elbow = right' then we find the elbow point to the right of the value that minimises the variance. Elbow points are found by drawing a line from the first point (for the left elbow) or the last point (for the right elbow) to x0, and finding the location on the curve of summed variances which maximises the distance to that line.

Value

Numeric value which splits the data.

Author(s)

Thomas J. Hardcastle

Examples

bimodalSeparator(c(rnorm(200, mean = c(5,7), sd = 1)))

baySeq documentation built on Nov. 8, 2020, 5:43 p.m.