To start the first step in the analysis and generate a tree set, you
have two options. You can either click on Tree
set
new tree set and follow the instructions to drag the
mouse to highlight the region of interest in the map
window. Alternatively, you can mark the region of interest in the map
window without using the Main Menu. Either step will open a pop-up
dialog box where all the necessary model parameters can be set. This
window asks for the beginning and the end of the studied region in bp
and the number of focal points that should be evenly distributed in
that region. After clicking on the button labeled More
options, the burn-in length for the MCMC in steps
of updates can be entered. Each step consists of 10,000 proposed
updates to the tree. Some discussion on what burn-in to choose here is
presented in section 8. Furthermore you are given the
option of performing a reduced burn-in for every MCMC after the
first. If this option is chosen, the MCMC does not use a random
starting tree for the second and all further focal points. Instead the
last tree that has been generated at the neighboring focal point is
used as the first tree for the new Markov Chain. The advantage of
using this option is that the burn-in for all but the first MCMC can
be substantially lower, reducing computation time. This gain is
maximized if the program is running on a single CPU. If the
tree-building process is distributed to multiple processors (see
7.2), each of these processors runs one
full-length MCMC as its first focal point and only subsequent Markov
Chains profit from the reduced burn-in. The disadvantage of this
option is that the tree sets at neighboring focal points are not
entirely independent, which may make it harder to spot aberrant
results as described in 8.2. See 8.1 for further
discussion of this option.
You can also enter the number of trees that should be sampled after burn-in. Each tree is sampled after one step of updates, consisting of 10,000 proposed changes to the tree. A good starting value for the number of sampled trees analysis is between 100 and 200 trees. Larger numbers become computationally very burdensome and offer only a limited advantage. Section 6.6 describes how more trees can be sampled if the initial set of trees is insufficient. Finally, the seed to initialize the random number generator can be set from this window. If you want to generate several independent runs of the treebuilding algorithm, make sure that each run has a different seed.
After this setup of the treebuilding process has been finished, a button Start running trees appears in the Tree Display. Clicking this button will start running the tree-building.