Configuring the Heuristic Preplacement
By default, EPA-ng performs placement of a sequence in two stages: first selecting
promising branches quickly (preplacement), then evaluating the selected branches in
greater detail.
EPA-ng currently offers three ways of selecting these candidates.
The default is the accumulated threshold method, in which branches are added to the set
of candidates until the sum of their LWR exceed a user specified threshold. The flag
controlling this mode is -g (or --dyn-heur ), with a default setting of 0.99999 ,
corresponding to a covered likelihood weight of 99.999%.
The second mode functions identically to the candidate selection mode in the original
implementation of the EPA in RAxML . Here again the branches are sorted by the LWR of
the placement of a sequence. Then, the top x% of the total number of branches are
selected into the set of candidates. Like in RAxML , this behavior is controlled via the -
G (or --fix-heur ) flag.
The third mode works identically to the baseball heuristic from pplacer, with default
settings (--strike-box 3.0, --max-strikes 6, --max-pitches 40) and is enabled using --baseball-heur flag.
Lastly, to disable the preplacement completely, you can simply supply the --no-heur flag.
Be warned however: doing so will be significantly more computationally demanding. Our
advice is to use the heuristic, as it sacrifices only insignificant amounts of accuracy for greatly improved speed.