Bayesian inference methods for parameter estimation : Implementation and benchmarking

Sammanfattning: In this study, three implementations of the nested sampling algorithm have been implemented, in the programming language Rust, compared and benchmarked. These variations consist of a classic version, closely resembling the first appearance of the algorithm, a version that produces samples from a single ellipsoid, and a version tha tuses multiple ellipsoids to generate its samples. These versions where compared to each other and counterparts from two Python libraries, Nestle and pymultinest. Testing the variations of the algorithms found that the multi ellipsoids sampler is the most versatile alternative and when comparing wall clock time, The Rust implementation of the multi ellipsoid sampler ran up to 79 times faster than its Nestle counterpart and up to 51 times faster than pymultinest. Running the rust implementations in parallel with eight threads proved to be slower in most examples, but in computationally difficult problems, the single ellipsoid sampler received a speedup of up to 3.50 while the multi ellipsoid sampler got a speedup of up to 1.86 when running benchmarks with eight threads rather than one.