secsse_ml: Maximum likehood estimation for (SecSSE)

View source: R/secsse_ml.R

secsse_mlR Documentation

Maximum likehood estimation for (SecSSE)

Description

Maximum likehood estimation under Several examined and concealed States-dependent Speciation and Extinction (SecSSE)

Usage

secsse_ml(
  phy,
  traits,
  num_concealed_states,
  idparslist,
  idparsopt,
  initparsopt,
  idparsfix,
  parsfix,
  cond = "proper_cond",
  root_state_weight = "proper_weights",
  sampling_fraction,
  tol = c(1e-04, 1e-05, 1e-07),
  maxiter = 1000 * round((1.25)^length(idparsopt)),
  optimmethod = "subplex",
  num_cycles = 1,
  loglik_penalty = 0,
  is_complete_tree = FALSE,
  verbose = (optimmethod == "simplex"),
  num_threads = 1,
  atol = 1e-08,
  rtol = 1e-07,
  method = "odeint::bulirsch_stoer"
)

Arguments

phy

phylogenetic tree of class phylo, rooted and with branch lengths.

traits

vector with trait states for each tip in the phylogeny. The order of the states must be the same as the tree tips. For help, see vignette("starting_secsse", package = "secsse").

num_concealed_states

number of concealed states, generally equivalent to the number of examined states in the dataset.

idparslist

overview of parameters and their values.

idparsopt

a numeric vector with the ID of parameters to be estimated.

initparsopt

a numeric vector with the initial guess of the parameters to be estimated.

idparsfix

a numeric vector with the ID of the fixed parameters.

parsfix

a numeric vector with the value of the fixed parameters.

cond

condition on the existence of a node root: "maddison_cond", "proper_cond" (default). For details, see vignette.

root_state_weight

the method to weigh the states: "maddison_weights", "proper_weights" (default) or "equal_weights". It can also be specified for the root state: the vector c(1, 0, 0) indicates state 1 was the root state.

sampling_fraction

vector that states the sampling proportion per trait state. It must have as many elements as there are trait states.

tol

A numeric vector with the maximum tolerance of the optimization algorithm. Default is c(1e-04, 1e-05, 1e-05).

maxiter

max number of iterations. Default is 1000 * round((1.25) ^ length(idparsopt)).

optimmethod

A string with method used for optimization. Default is "subplex". Alternative is "simplex" and it shouldn't be used in normal conditions (only for debugging). Both are called from DDD::optimizer(), simplex is implemented natively in DDD, while subplex is ultimately called from subplex::subplex().

num_cycles

Number of cycles of the optimization. When set to Inf, the optimization will be repeated until the result is, within the tolerance, equal to the starting values, with a maximum of 10 cycles.

loglik_penalty

the size of the penalty for all parameters; default is 0 (no penalty).

is_complete_tree

logical specifying whether or not a tree with all its extinct species is provided. If set to TRUE, it also assumes that all all extinct lineages are present on the tree. Defaults to FALSE.

verbose

sets verbose output; default is TRUE when optimmethod is "simplex". If optimmethod is set to "simplex", then even if set to FALSE, optimizer output will be shown.

num_threads

number of threads to be used. Default is one thread.

atol

A numeric specifying the absolute tolerance of integration.

rtol

A numeric specifying the relative tolerance of integration.

method

integration method used, available are: "odeint::runge_kutta_cash_karp54", "odeint::runge_kutta_fehlberg78", "odeint::runge_kutta_dopri5", "odeint::bulirsch_stoer" and "odeint::runge_kutta4". Default method is: "odeint::bulirsch_stoer".

Value

Parameter estimated and maximum likelihood

Examples

# Example of how to set the arguments for a ML search.
library(secsse)
library(DDD)
set.seed(13)
# lambdas for 0A and 1A and 2A are the same but need to be estimated
# mus are fixed to
# the transition rates are constrained to be equal and fixed 0.01
phylotree <- ape::rcoal(31, tip.label = 1:31)
traits <-  sample(c(0,1,2), ape::Ntip(phylotree),replace=TRUE)#get some traits
num_concealed_states<-3
idparslist <- id_paramPos(traits, num_concealed_states)
idparslist[[1]][c(1,4,7)] <- 1
idparslist[[1]][c(2,5,8)] <- 2
idparslist[[1]][c(3,6,9)] <- 3
idparslist[[2]][]<-4
masterBlock <- matrix(5,ncol = 3,nrow = 3,byrow = TRUE)
diag(masterBlock) <- NA
diff.conceal <- FALSE
idparslist[[3]] <- q_doubletrans(traits,masterBlock,diff.conceal)
startingpoint <- DDD::bd_ML(brts = ape::branching.times(phylotree))
intGuessLamba <- startingpoint$lambda0
intGuessMu <- startingpoint$mu0
idparsopt <- c(1,2,3,5)
initparsopt <- c(rep(intGuessLamba,3),rep((intGuessLamba/5),1))
idparsfix <- c(0,4)
parsfix <- c(0,0)
tol <- c(1e-02, 1e-03, 1e-04)
maxiter <- 1000 * round((1.25)^length(idparsopt))
optimmethod <- 'subplex'
cond <- 'proper_cond'
root_state_weight <- 'proper_weights'
sampling_fraction <- c(1,1,1)
model<-secsse_ml(
phylotree,
traits,
num_concealed_states,
idparslist,
idparsopt,
initparsopt,
idparsfix,
parsfix,
cond,
root_state_weight,
sampling_fraction,
tol,
maxiter,
optimmethod,
num_cycles = 1,
verbose = FALSE)
# model$ML
# [1] -16.04127

secsse documentation built on Oct. 22, 2023, 1:13 a.m.