abcpp is a lightweight embeddable C++ library for Approximate Bayesian Computation, with R and Python wrappers. The C++ library is the only algorithm implementation; R and Python are thin frontends.

C++ Usage

abcpp can be embedded in another CMake project in the same style as small algorithm libraries such as NLopt, using FetchContent_Declare to call abcpp as the algorithm backend:

include(FetchContent)
FetchContent_Declare(
  abcpp
  GIT_REPOSITORY https://github.com/yuki-961004/abcpp.git
  GIT_TAG main
  GIT_SHALLOW TRUE
)
FetchContent_MakeAvailable(abcpp)
target_link_libraries(my_target PRIVATE abcpp::abcpp)

The primary C++ API is NLopt-like:

#include <abcpp/abcpp.hpp>
abcpp::opt opt;
abcpp::result fit = opt
    .set_target(target)
    .set_params(params)
    .set_sumstats(sumstats)
    .set_method(abcpp::method::neuralnet)
    .set_tol(0.01)
    .set_nnet_sizenet(8)
    .run();

R Usage

Install from CRAN:

install.packages("abcpp")

The R interface uses four inputs: target, params, sumstats, and control.

library(abcpp)
result <- abcpp::abc(
  target = <target_summary_vector_or_matrix>,
  params = <parameter_vector_or_matrix>,
  sumstats = <simulated_summary_vector_or_matrix>,
  control = list(
    method = "rejection",
    tol = <tolerance_between_0_and_1>
  )
)
summary(result)

Python Usage

Install from PyPI:

pip install abcpp

The Python interface mirrors the R interface:

import abcpp
result = abcpp.abc(
    target=<target_summary_vector_or_matrix>,
    params=<parameter_vector_or_matrix>,
    sumstats=<simulated_summary_vector_or_matrix>,
    control={
        "method": "rejection",
        "tol": <tolerance_between_0_and_1>,
    },
)
abcpp.summary(result)