dist_match: Distribution Matching for Source and Reference Datasets

In DMTL: Tools for Applying Distribution Mapping Based Transfer Learning

Description

This function matches a source distribution to a given reference distribution such that the data in the source space can effectively be transferred to the reference space i.e. domain transfer via distribution matching.

Usage

dist_match(
  src,
  ref,
  src_cdf,
  ref_cdf,
  lims,
  density = FALSE,
  samples = 1e+06,
  seed = NULL
)

Arguments

src	Vector containing the source data to be matched.
ref	Vector containing the reference data to estimate the reference distribution for matching.
src_cdf	Vector containing source distribution values. If missing, these values are estimated from the source data using estimate_cdf().
ref_cdf	Vector containing reference distribution values. If missing, these values are estimated from the reference data using estimate_cdf().
lims	Vector providing the range of the knot values for mapping. If missing, these values are estimated from the reference data.
density	Flag for using kernel density estimates for matching instead of histogram counts. Defaults to False.
samples	Sample size for estimating distributions if src_cdf and/or ref_cdf are missing. Defaults to 1e6.
seed	Seed for random number generator (for reproducible outcomes). Defaults to NULL.

Value

A vector containing the matched values corresponding to src.

Examples

set.seed(7531)
x1 <- rnorm(100, 0.2, 0.6)
x2 <- runif(200)
matched <- dist_match(src = x1, ref = x2, lims = c(0, 1))

## Plot histograms...
opar <- par(mfrow = c(1, 3))
hist(x1);    hist(x2);    hist(matched)
par(opar)              # Reset par

DMTL documentation built on Feb. 18, 2021, 5:06 p.m.