Overlap quantification for LPS object

Description

Quantify the overlap between gaussian distributions of the two group scores, to assess model efficiency (best models should not overlap, to prevent from false discovery).

Usage

1
  OVL(means, sds, cutoff=1e-4, n=1e4)

Arguments

means

Numeric vector of two values, the means of the gaussian distributions.

sds

Numeric vector of two values, the standard deviations of the gaussian distributions.

cutoff

Single numeric value, minimal quantile for integration range definition (distributions will be considered between their cutoff and 1 - cutoff quantiles only). The lesser it is, the more precise the returned value will be.

n

Single integer value, the amount of equi-distant points to use for the computation. The greater it is, the more precise the returned value will be.

Value

Returns the proportion of the overlap between the two gaussian distributions N1 and N2, i.e. min(N1, N2) / (N1 + N2).

Author(s)

Sylvain Mareschal

See Also

LPS-class, LPS, link{dnorm}

Examples

1
2
3
4
5
6
7
8
  # Full overlap between identical distributions
  OVL(c(0,0), c(1,1))
  
  # Increasing shift
  OVL(c(0,1), c(1,1))
  OVL(c(0,2), c(1,1))
  OVL(c(0,3), c(1,1))
  OVL(c(0,10), c(1,1))