Module for Approximate Bayesian Computation
Approximate Bayesian Computation (ABC) algorithms, also called likelihood free inference techniques, are a family of methods that can render virtually impossible models to feasible scale. Additionally, the ABC approach has proven to be successful over likelihood based methods in several instances and is conceptually linked to recent and very innovative Deep Learning developments such as Generative Adversary Nets. We propose to implement a module for ABC in PyMC3, specifically Sequential Monte Carlo-ABC (SMC-ABC). Our work will signify a meaningful increase in the spectrum of models that PyMC3 will be able to perform.