We continue using material from the short course Simulation-Based Inference for Epidemiological Dynamics (SBIED). We review likelihood-based inference, and introduce the particle filter (also known as sequential Monte Carlo) which enables likelihood evaluation for general POMP models. We use the pfilter() implementation of particle filtering in pomp to investigate the likelihood surface for a Susceptible-Infected-Recovered (SIR) model.
Sections I and II review material in Chapter 10, from a slightly different perspective, and you may want to watch this at 1.5 x speed if you are already comfortable with the material. Section III introduces the particle filter algorithm. Section IV reviews the likelihood-based inference topics of Chapter 5, in the context of POMP models. Sections V and VI look at the geometry of the likelihood for an SIR model. Section VII looks deeper at likelihood inference - there is overlap with Chapter 5 but also some important new ideas.
| Slides | ||
| Annotated slides | ||
| Notes | ||
| R script | R | |
| Recording, Chapter 13, Section III | Computing the likelihood | (27 mins) |
| Recording, Chapter 13, Section IV | Review of likelihood-based inference | (12 mins) |
| Recording, Chapter 13, Sections V,VI | Geometry of the likelihood function | (31 mins) |
| Recording, Chapter 13, Section VII | From likelihood evaluation toward inference | (17 mins) |
Back to course homepage
Acknowledgements
Source code for these notes