AbcRanger

a fast and scalable random forest library for ABC model choice and parameter estimation

A brief introduction to the AbcRanger library and bayesian methologies for ABC model choice and parameter estimation.
Published

January 13, 2020

Abstract

The AbcRanger library (https://github.com/diyabc/abcranger) provides methodologies for abc model choice and parameter estimation based on fast and scalable random forest implementation, tuned to handle large and high dimensional datasets.

Using a modified C++ implementation of state-of-the-art random forests (https://github.com/imbs-hl/ranger), we do not store deep decision trees in memory but process them by batches in parallel

The library was first intended to be used with a population genetics ABC framework (https://github.com/diyabc/diyabc), but has been generalized to any ABC reference table generator. We focused on memory and thread scalability, ease of use (minimal hyperparameter set). R and python interfaces are provided.