rmcminds / cophylogeny_simulations Goto Github PK

modeling shifts in variance ultimately doesn't model the generative behavior of codiversification. the generative behavior is discrete: once codiv starts, it is a binary question at each generation: which offspring receive which descendant. playing with variance is weird partially because the inheritance of the increased variance is tied to the location effect: variance goes to zero in one offshoot, but variance increases in the other.

problem with codiv in stan is modeling latent discrete states. need to marginalize over them, so math is not as naive as simply declaring latent continuous variables and using them in the model directly. fortunately, modeling latent discrete states is exactly what is done with a hidden markov model, which has a guide to stan implementation.

https://mc-stan.org/docs/2_29/stan-users-guide/hmms.html

https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1008030

https://aclanthology.org/W13-2263.pdf

https://link.springer.com/content/pdf/10.1007%2F978-0-387-70898-0_8

https://link.springer.com/chapter/10.1007/978-0-387-70898-0_8

https://hal.inria.fr/hal-00830078/document