check_BioGeoBEARS_run(inputs, allow_huge_ranges = FALSE)cladoRcpp's
numstates_from_numareas function
to calculate the size of the state space ahead of time,
and links therein to see how the number of states scales
with areas (2^number of areas, in an unconstrained
analysis), how the size of the transition matrix you will
be exponentiating scales (size = numstates * numstates),
and the size of the
ancestor/left-descendant/right-descendant cladogenesis
matrix scales (numstates * numstates * numstates). At
500 states, this is 500^3 = 125,000,000 combinations of
ancestor/left/right to check at every cladogenesis event,
although cladoRcpp's tricks
speed this up substantially.TRUE if no errors found; otherwise a stop() is
called.
- Trees with negative branchlengths (as produced sometimes by e.g. BEAST MCC consensus trees (MCC = majority clade consensus). These trees are always fully resolved, but the median node heights can sometimes be behind the node position in the tree. Users should fix this manually, pathological results or crashes will result otherwise.
- Trees with polytomies. BioGeoBEARS (and
LAGRANGE, and DIVA) assume a model where
lineages bifurcate, and never multifurcate. Users can
convert multifurcating trees to bifurcating trees with
APE's multi2di (they will have
to decide what branchlength to use for the new branches;
it should be small, but bigger than the minimum
branchlength used to identify fossils hooks (as hooks are
considered to be anagenetic members of a lineage, and
thus are connected to the tree without a cladogenesis
event invoked). Users can then run their analysis several
times on differently-resolved trees.
NOTE: After the above correction, users may wish to
correct the tip branchlengths (or make some other
adjustment) so that all the tips are at age 0 my before
present, as in an ultrametric tree. (However, note that
trees with fossil tips are not ultrametric according to
APE's is.ultrametric, even though they
are time-scaled. To make living (nonfossil) tips line up
to zero, see average_tr_tips or the
(different!) . They should be used with care.
Alternately, a small amount of error in tip heights will
make very little difference in the likelihood
calculations (e.g. if some tips are 0.1 my too high, but
the tree spans 200 my), which would be an argument for
not requiring perfection after the (crucial) corrections
of negative branchlengths, zero-branchlengths, and
polytomies have been made.
- Check for an absurdly large number of states. I've set
the limit at 500 (it starts getting slow around 200),
users can override with allow_huge_ranges=TRUE.
- Geography tipranges files should have same number of area labels as columns.
- Geography tipranges files should have same number of
taxa as the tree, and with the (exact!!) same names.
This can be the source of many headaches, as different
programs (Mesquite, etc.) treat spaces, periods,
etc. in different ways, and re-write tipnames
with/without quotes, underscores, etc.; and in my
experience, my biologist colleagues find it very
difficult to guarantee that the tipnames in their tree
and their data tables will match exactly. The SAFEST
approach is to NEVER use these characters in tipnames or
table names: space, comma, semicolon, dashes,
parentheses, brackets, apostrophes or quote marks, or
periods. Use ONLY letters, numbers, and underscores (_).
When plotting trees, APE automatically reads
underscores as spaces, which is nice for display.
- There must be the same or more timeperiods than the other stratified items (distances matrices, etc.)
Matzke_2012_IBS
average_tr_tips,