Particle filter code for research in static and time dependent parameter estimation. Derived from:
C. Gentner, S. Zhang, and T. Jost, “Log-PF: particle filtering in logarithm domain,” Journal of Electrical and Computer Engineering, vol. 2018, Article ID 5763461, 11 pages, 2018.
D. Calvetti, A. Hoover, J. Rose, and E. Somersalo. Bayesian particle filter algorithm for learning epidemic dynamics. Inverse Problems, 2021
The code is structured in a class hierarchy, abstract base classes(ABC) are used to template implementations of the basic algorithm.
Abstract/Algorithm: This ABC is the template for the entrypoint of the algorithm, it has 5 fields, 3 of which all also ABC's templating subfunctionality required for running the algorithm.
Abstract/Perturb: This ABC is one of the fields of Algorithm, it templates a class which perturbs the particles according to some distribution. This class templates only one method, "randomly_perturb", and one field, hyperparameters, which is a dictionary of hyperparameters for the perturbation.
Abstract/Resampler: This ABC is one of the fields of Algorithm, it performs the resampling step of the particle filter and templates two functions, "compute_weights" to compute the particle weights using the likelihood function, and "resample" to perform the actual resampling.