AS.select {abctools}

Summary statistics selection using approximate sufficiency.

Description

This function uses approximate sufficiency to assess subsets of summary statistics for ABC inference.

Usage

AS.select(obs, param, sumstats, obspar=NULL, abcmethod=abc, 
grid=10, inturn=TRUE, limit=ncol(sumstats), allow.none=FALSE, 
do.err=FALSE, final.dens=FALSE, errfn=rsse, trace=TRUE, ...)

Arguments

obs

(matrix of) observed summary statistics.

param

matrix of simulated model parameter values.

sumstats

matrix of simulated summary statistics.

obspar

optional observed parameters (for use to assess simulation performance).

abcmethod

a function to perform ABC inference, e.g. the abc function from package abc.

grid

the number of bins into which to divide the posterior sample for the approximate sufficiency calculation.

inturn

a boolean value indicating whether "bad" statistics should be dropped and tested sequentially (inturn=TRUE) or all at the end (inturn=FALSE).

limit

an optional integer indicating whether to limit summary selection to subsets of a maximum size.

allow.none

a boolean values indicating whether an empty subset of statistics is considered in the selection procedure.

do.err

a boolean value indicating whether the simulation error should be returned. Note: if do.err=TRUE, obspar must be supplied.

final.dens

a boolean value indicating whether the posterior sample should be returned.

errfn

an error function to assess ABC inference performance.

trace

whether to show progress messages.

...

any other optional arguments to the ABC inference procedure (e.g. arguments to the abc function).

Details

The summary selection procedure works by sequentially testing randomly chosen statistics for inclusion, using the ratio of ABC posterior samples to determine whether a statistic is added. Since adding a statistic may result in a suboptimal subset of summaries, the included statistics are then individually dropped and retested, to determine whether a smaller subset of statistics is equally / more informative than the accepted set of statistics.

Value

A list with the following components:

best

the final subset of included statistics.

err

simulation error (if obspar is supplied and do.err=TRUE).

post.sample

an array of dimension nacc x npar x ndatasets giving the posterior sample for each observed dataset. Not returned if final.dens=FALSE.

Note

The approximate sufficiency techniques described here are only suitable for single parameters only.

Author(s)

Matt Nunes

References

Blum, M. G. B, Nunes, M. A., Prangle, D. and Sisson, S. A. (2013) A comparative review of dimension reduction methods in approximate Bayesian computation. Stat. Sci. 28, Issue 2, 189–208.

Joyce, P. and P. Marjoram (2008) Approximately sufficient statistics and Bayesian computation. Stat. Appl. Gen. Mol. Biol. 7 Article 26.

Nunes, M. A. and Prangle, D. (2016) abctools: an R package for tuning approximate Bayesian computation analyses. The R Journal 7, Issue 2, 189–205.

See Also

AS.test

Examples

# load example data:

data(coal)
data(coalobs)

param<-coal[,2]
simstats<-coal[,4:6]

# use matrix below just in case to preserve dimensions.

obsstats<-matrix(coalobs[1,4:6],nrow=1)

# example of AS.select:

## Not run: 
tmp <-AS.select(obsstats, param, simstats, tol=.1, method="neuralnet",
nument=5, allow.none=FALSE, inturn=TRUE)

tmp$best

## End(Not run)