Approximate Bayesian Computations (ABC) are a Monte Carlo technique to perform approximate parameter inference when the likelihood term cannot be easily evaluated. ABC proceeds by summarising the data, simulating from the model, comparing simulated summaries to observed summaries with a distance function, and accepting the simulated summaries if they do not differ from the observed summaries by more than a user-defined tolerance parameter. These steps are repeated through many Monte Carlo iterations to obtain an approximation to the true posterior density of the model parameters. The process by which precise ABC tolerances and ABC distance functions can be obtained is often referred to as ABC calibrations. These calibrations make use of decision theoretic arguments to construct the ABC accept/reject step, so that the ABC accept/reject step enjoys certain desirable properties.
The abc.star
R package implements ABC calibration routines for the most commonly occurring scenarios.
The corresponding paper is now on arxiv http://arxiv.org/abs/1305.4283
If you'd like to add calibration routines for different equivalence tests to this library, please let us know and we are happy to give you write access to the github repository
The easiest way to install abc.star
is to use the devtools
package:
# install.packages("devtools")
library(devtools)
install_github("olli0601/abc.star")
To work with this R
package:
fire up R
, and type
library(help=abc.star)
ABC calibration routines for particular testing scenarios. These can be combined like lego blocks to build your calibrated ABC algorithm.
mutost.*
vartest.*
ratetest.*
ztest.*
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.