## Even faster version of rcpp_calc_anclikes_sp

### Description

This function improves on rcpp_calc_anclikes_sp and rcpp_calc_anclikes_sp_COOprobs. In addition to the compressed COO-like storage format, the internal C++ code here explicitly enumerates the allowed transitions, rather than searching through every possibility and testing whether or not it is allowed. This appears to scale well to very large state spaces.

### Usage

rcpp_calc_anclikes_sp_COOweights_faster(Rcpp_leftprobs, Rcpp_rightprobs, l,
s = 1, v = 1, j = 0, y = 1, dmat = NULL, maxent01s = NULL,
maxent01v = NULL, maxent01j = NULL, maxent01y = NULL,
max_minsize_as_function_of_ancsize = NULL, printmat = TRUE,
m = NULL, m_null_range = TRUE, jts_matrix = NULL)


### Arguments

 Rcpp_leftprobs Probabilities of the states at the base of the left descendant branch Rcpp_rightprobs Probabilities of the states at the base of the right descendant branch l List of state indices (0-based) s Relative weight of sympatric "subset" speciation. Default s=1 mimics LAGRANGE model. v Relative weight of vicariant speciation. Default v=1 mimics LAGRANGE model. j Relative weight of "founder event speciation"/jump speciation. Default j=0 mimics LAGRANGE model. y Relative weight of fully sympatric speciation (range-copying). Default y=1 mimics LAGRANGE model. dmat If given, a matrix of rank numareas giving multipliers for the probability of each dispersal event between areas. Default NULL, which sets every cell of the dmat matrix to value 1. Users may construct their own parameterized dmat (for example, making dmat a function of distance) for inclusion in ML or Bayesian analyses. maxent01s Matrix giving the relative weight of each possible descendant rangesize for the smaller range, for a given ancestral rangesize, for a subset-sympatric speciation event. Default is NULL, which means the script will set up the LAGRANGE model (one descendent always has range size 1). maxent01v Matrix giving the relative weight of each possible descendant rangesize for the smaller range, for a given ancestral rangesize, for a vicariance speciation event. Default is NULL, which means the script will set up the LAGRANGE model (one descendent always has range size 1). maxent01j Matrix giving the relative weight of each possible descendant rangesize for the smaller range, for a given ancestral rangesize, for a founder-event speciation event. Default is NULL, which means the script will set up the LAGRANGE model (one descendent always has range size 1). maxent01y Matrix giving the relative weight of each possible descendant rangesize for the smaller range, for a given ancestral rangesize, for a full-sympatric (range-copying) speciation event. Default is NULL, which means the script will set up the LAGRANGE model (one descendent always has range size 1). max_minsize_as_function_of_ancsize If given, any state with a range larger that this value will be given a probability of zero (for the branch with the smaller rangesize). This means that not every possible combination of ranges has to be checked, which can get very slow for large state spaces. printmat Should the probability matrix output be printed to screen? (useful for debugging, but can be dramatically slow in R.app for some reason for even moderate numbers of states; perhaps overrunning the line length...) m This is a vector of rate/weight multipliers for dispersal, conditional on the values of some (non-biogeographical) trait. For example, one might hypothesize that flight/flightlessness effects dispersal probability, and manually put a multiplier of 0.001 on the flightlessness state. Or, one might attempt to estimate this. The strategy used in cladoRcpp is to expand the default cladogenetic rate matrix by length(m) times. I.e., if m is not NULL, then loop through the values of m and apply the multipliers to d (and j, and a) events. Default is NULL. m_null_range Is the null range included in the state space in the general analysis? (The function needs to know this, when there are traits, to index the state space correctly.) jts_matrix A numtraits x numtraits matrix containing the proportions for trait transitions during j events. E.g., for a sudden switch from trait 1 (flight) to trait 2 (flightlessness) during a jump event.

### Details

This should be faster, i.e. by look for each type of event individually.

Returns results as 4 columns: ancestral index, left index, right index, conditional probability given ancestral states (assuming likelihood of descendants is 1). Indexes are 0-based.

Keep in mind that cladogenesis matrices exclude the null state (a range of 0 areas), so if your states list starts with the null range (as is typical/default in DEC-style models) then to get the R 1-based state indices requires e.g. COO_weights_columnar[[1]] + 2.

When the calculation is run at each node in the tree, all that is required is one pass through the COO-like array, with the downpassed probabilities of the states on the left and right branches multiplied by the probability column.

### Value

COO_weights_columnar Transition weights matrix in COO-like format as 4 columns: ancestral index, left index, right index, and weight of the specified transition. Indexes are 0-based. Keep in mind that cladogenesis matrices exclude the null state (a range of 0 areas), so if your states list starts with the null range (as is typical/default in DEC-style models) then to get the R 1-based state indices requires e.g. COO_weights_columnar[[1]] + 2.

Dividing the weights by the sum of the weights for a particular ancestral state yields the conditional probabilities of each transition at the speciation event. (assuming likelihood of descendants is 1).

### Author(s)

Nicholas Matzke matzke@berkeley.edu

rcpp_calc_anclikes_sp, rcpp_calc_anclikes_sp_COOprobs, rcpp_calc_anclikes_sp_COOweights_faster

rcpp_calc_anclikes_sp #bibliography /Dropbox/_njm/__packages/cladoRcpp_setup/cladoRcpp_refs.bib @cite Matzke_2013 @cite Matzke_2014 @cite ReeSmith2008

### Examples

# For the basic logic of a probablistic cladogenesis model, see
?rcpp_calc_anclikes_sp

# For examples of running the functions, see the comparison of all functions at: