Computing

Research

The failure of the Anoles data set to robustly select the more common model can easily be explained by the likelihood distribution: Not surprising since this dataset consists of only 23 taxa. The Labrid tree of 114 taxa clearly resolves the difference between the different models. Bootstrapping should be a more reliable guide than comparing AIC scores.

image
image

Compare to the Labrid data set:

image
image

These use 200 bootstrap simulations, repeating with 2000 for better resolution.


### Significance

Breakthrough! This approach has far greater implications than I first realized. This provides an inverse approach which solves the hardest part of my original problem, and then tests that under the phylogenetic context. Solving the partition problem independent of the phylogeny, reconstructing the ancestral states as a purely discrete problem, and then scoring the resulting continuous time model on the phylogeny is likely to be far more effective, efficient, and robust, then the direct calculation of the joint probability that I’ve been pursuing. While the three steps: partitioning, discrete inference, and continuous inference; all need refining, I think I’m very close to my first implementation of the full problem!


### Next Steps

  • Compare likelihood scores under different levels of partitioning.
  • Parametric bootstrapping of partitioning
  • look at parametric bootstrap of parameter values inferred as well as the model likelihood.
  • Stochastic Painting from ancestral state reconstruction on discrete traits
  • Functionalize the three steps to inference


### Other Progress

  • Working on implementing an ouch2ape function to get ouch trees back into ape.
  • Begun functionalizing partitioning, discrete and continuous inference. Converting between formats without getting data misaligned is remarkably complicated.

From the git log

commit ddf6db6fca626c8f6ed31f4ab2a7cb78ed70f226 Author: Carl Boettiger Date: Mon Mar 15 09:40:08 2010 -0700

   Updated doxygen comments

commit 17dd3775004741c9c590828b498dacd75c5bfc49 Author: Carl Boettiger Date: Mon Mar 15 09:39:34 2010 -0700

   added license and multimin fn

commit 02b9bfc02cb830297fc7fad81cfcd05598d482f1 Author: Carl Boettiger Date: Mon Mar 15 09:22:21 2010 -0700

   mcmc is basics now implemented.

commit 220ccae12fbbdf78cd5e082555c04b99af92c4af Author: Carl Boettiger Date: Sun Mar 14 20:51:50 2010 -0700

   adding the mcmc routine

commit 94642c8d406fb0d6e8970dc31cfd5304dda2c007 Author: Carl Boettiger Date: Sun Mar 14 15:33:39 2010 -0700

   first commit