bomeara / selac Goto Github PK

Right now have a mismatch of documentation.

I am documenting this error here until someone (including myself) fixes this error.

Sometimes, the following error is given when the first likelihood is being calculated...

1: In mclapply(1:nsites.unique, MultiCoreLikelihoodBySite, mc.cores = n.cores.by.gene.by.site) :
  scheduled cores 2 encountered errors in user code, all values of the jobs will be affected

Exploration of the error shows that it is somehow related to the input alignment, which in my case was a codon alignment. This can be resolved by deleting codons sites for which all taxa have an incomplete codon.

I keep getting the error

Error in data[, tree$tip.label] : subscript out of bounds

and I am not sure why

Hessian slow. Adaptive sampling less slow. Maybe root finding? Maybe another algorithm? What cutoff for adaptive sampling (∆lnL=2?)

Something to make a tree and codon data, convert to codon numeric codes, make sure runs with rayDisc to get likelihood

@dominicev's results suggest that this might be more efficient. Could do a later more intense search on the best.

R currently gives this message Error in data[, tree$tip.label] : subscript out of bounds

When someone inputs seqs with stop codons, maybe delete from that point on in that gene for all taxa? Or just stop with an error?

Code is currently like

while(number.of.current.restarts < (max.restarts+1)){
  optimize edge lengths (one round of nloptr, in parallel) #line 3566
  optimize parameters (one round of nloptr, in parallel) #line 3577
  while (lik.diff != 0 & iteration.number<7){ #line 3592
    optimize edge lengths (one round of nloptr, in parallel) #line 3599
    optimize parameters (one round of nloptr, in parallel) #line 3613   
    lik.diff <- round(abs(current.likelihood-results.final$objective), 8) #line 3451
  }
}

but the 8 for round is hardcoded -- expose this as a fn argument.

Look at RCpp

opts <- list("algorithm" = "NLOPT_LN_SBPLX", "maxeval" = max.evals, "ftol_rel" = max.tol) in SelacOptimize: expose algorithm as an argument.