Figure 5
Approximate Bayesian computation with population Monte Carlo (ABC-PMC). (a) Given the observed data. (b) Particles are generated from a proposal distribution and data is simulated for each particle. For each particle, the Wasserstein distance is measured between the simulated data and the observed data. (c) This is repeated until nParticles samples are generated with Wasserstein distance within a tolerance \(\epsilon\). (d) A new proposal distribution is generated by a weighted kernel density estimate on the accepted particles, with a weighting based on importance sampling principles. A new tolerance is set based upon a proportion of survivalFraction particles with the smallest distances found in this time step. This is repeated for a given number of generations. The final successful particles are used to generate an approximation of the posterior distribution using a weighted kernel density estimate. Figure adapted in part from19 and21.
