Multi-layered linear stochastic differential equation inference

This page presents the supplementary material and software used for multi-layered analysis using vectorial linear stochastic differential equations. The reason for using stochastic differential equation is to facilitate studying irregularly spaced observations like those we find in our study dataset of micro-fossils. This work was performed in conjuction with doing the research for and writing the manuscript "Modelling Evolution by Linear Stochastic Differential Equations with Latent Layers" by T. Reitan, T. Schweder and J. Henderiks. A short powerpoint presentation can be found here, with a PDF version here. (Made for the Seventh Conference on Bayesian Inference in Stochastic Processes in Madrid, Spain, BISP7.) A simpler presentation of the basic modelling framework and how it connects to the application can be found here. A how-to concerning the program usage, with a short intro to continuous time modelling is found here.

Originally, this was used for modelling phenotypic evolution towards a optima that itself could be a stochastic process depending responding to layers below it (climate or primary optimum). However, it may well be that the method can be used for other applications as well, with a little tweaking. The program found here so far, was geared towards solving the specific problem of Coccolithus microfossils though, but has lately been made to solve more general linear continuous time process inference problems.

The program (as well as the previously used underlying program library Hydrasub) is licenced as LGPL-based, meaning that anyone can modify and use it for other purposes than the original intent.

The latest version of the analysis program and R package uses only the Lapack library. Older versions use the Hydrasub library for support. This library in turn uses GSL (Gnu Scientific Library). Plotting programs, such as 'vvgraph' and 'histogramme' are used for displaying the results. PS: If plotting in the program (rather than the R package) is to be used, then even the latest version requires the 'vvgraph' and 'histogramme' programs from the Hydrasub library. However, there are options for not doing plots and for sending plots to file instead.



Up to my home page.
Trond Reitan
Last modified: Wed Nov 23 12:00:00 CET 2017