Automatic pipelined version of the Lancaster spatial stochastic SEIR model. In this release we have:

  1. Used ruffus to create complex pipeline of interacting Python modules, separating out data assembly, inference, prediction, and summaries.

  2. Bumped our TensorFlow and TensorFlow Probability versions to 2.4 and 0.12.1 respectively.

  3. Separated out TensorFlow Probability extensions into a partner library gemlib.

  4. Added many, many summary steps to provide incidence, prevalence, Rt, and within/between-LAD PAF for all LADs in the United Kingdom of Britain and Northern Ireland.

Roadmap/todo:

  1. Improve on time-varying infection rate using a Brownian motion model instead of 2-week varying Gaussian process;
  2. Hidden Markov-model implementation to account for test case under-reporting and allow more flexibility to adjust to changing diagnostic test sensitivity (e.g. PCR vs LFT) and differential testing effort.
  3. Vaccination implementation.