fullsearch: Search s for the best parameter

In danjlawson/robustarchitecture: Estimate Genomic Architecture parameters using robust statistics

Description

Optimize lossfunctionp. For basic usage, just give dat and it should do everything else for you. You sometimes have to change thresholds to fit the shape of your data; check this with architectureplot.

This is a 1-D non-convex optimization problem. The first challenge is that many parameters don't have a finite loss, so we first do a linesearch to identify a valid range.

We then either do a sequence of linesearches to arrive at a region, or go straight for a convex optimizer ("optimize" or "Brent" from optim). Line searches are slower but cope better with small datasets.

Usage

fullsearch(
  dat,
  bins = defaultbins(),
  thresholds = c(2.5, 3),
  sigma = NULL,
  method = c("linesearch", "optimize"),
  nlinesearch = 2,
  nline = 21,
  range = c(-2, 2)
)

Arguments

dat	A data frame of (f,beta) pairs
bins	(Default: defaultbins) the histogram bins to use
thresholds	(Default :defaultthresholds) the tail areas to compare (in s.d units).
sigma	(Default: NULL, meaning use the best value) Genomic architecture sigma. Don't change unless you know what you are doing...
method	(Default: c("linesearch","optimize")) The set of all methods to use (out of "linesearch", "optimize" and "Brent"). Good values are c("linesearch","optimize"), which tries to find global maxima, and "Brent" which tries to find a local maxima only, but is slightly faster. Only specify up to one of "Brent" and "optimize".
nlinesearch	(Defaul: 1) Number of linesearch parameters to evaluate, controlling the resolution near the edges of the valid range.
nline	(Defaul: 21) Number of linesearch parameters to evaluate in .
range	(Default: c(-2,2)) The lower and upper bounds provided to linesearch to find a region that returns a finite score.

Value

A list as returned by the optimiser, with additionally dat, thresholds, bins, range and sigma as provided, and s as inferred.

See Also

linesearch to explore a range of s, or bootstrap to generate a distribution of estimates of s.

danjlawson/robustarchitecture documentation built on Dec. 19, 2021, 8:09 p.m.