User:Carl Boettiger/Notebook/Comparative Phylogenetics/2010/04/05

From OpenWetWare
Jump to: navigation, search
Owwnotebook icon.png Comparative Phylogenetics Report.pngMain project page
Resultset previous.pngPrevious entry      Next entryResultset next.png

Research

  • Bootstrapping procedures should have the option to refit the painting. Currently use the fixed painting but refit parameters.
  • wrote the function peaks_update() which update the painting as well as the OU peak parameters.
20 replicates, bootstraps the painting step as well

Computing / Package

  • LR_bootstrap and model_bootstrap should be the same function call with an optional second model for the LR bootstrapping. Similarly the LR_bootstrap_all and model_bootstrap_all() should be the same function call. Will simplify documentation, error handling, redundancy etc.
    • Will need to combine the return object types: model_boots object or an LR_boots object though. Will need an indicator for their type, and combine a plot function that can handle both.
  • The fit_k_peaks can't actually take an ouchtree, though it will try, due to the strange behavior of the ape package in tree pruning discussed earlier. Will eventually need a better work around than writing out and reading back in the nexus file. Should post this issue to r-sig-phylo.
  • The refit painting bootstrap needs the original ape-format (phylo class) tree, due to the above problem. It's been added as an argument to the bootstrapping calls.

Troubleshooting

  • The partitions function may assign all points to only three regimes even though it is given four regimes to fit to. This has been causing a rather subtle bug in bootstrapping the ou4 model in the labrid dataset. Trying to address by modifying the function infer_painting(), a function child of fit_k_peaks() function, to calculate k as the maximum value in the factor list, rather than the length of the factor list. Still leads to other errors...
  • also tried modifying so that factor matching is explicit, lines 54, 55 of infer_niches.R Still get NAs in regime inference when its failed to use all the peaks.
  • Ways to force utilizing all peaks??

Speed concerns

  • Refitting is much slower. Will need to explore parallelizing each of the bootstrapping fits. Should also perform profiling and start moving slowest parts into C code. Will also need to avoid any redundant bootstrapping.
  • Shouldn't refit the true model in each row. Should be able to reuse the simulated data and the fit of the true model, and just refit each comparison model to that simulated dataset. Doesn't seem fit naturally in the framework of the bootstrap package, however.

Profiling Results

  • Profiling with Rprof(), summaryRprof()
  • fit_k_peaks() is 90% of computational time
    • it's member function, fit_k_partitions() is 65% total time
    • it's member function, infer_painting() is 10% of total time
    • optim function, called by fit_k_partitions() (also called by hansen) is 71% of total time

Misc / Notes

  • Following phylogenetics discussion group Friday, Samantha passed on new suggestions to Brian, now on the planned updates for Brownie.

Talks

  • Need to prepare Evolution Discussion Group talk for next week.
  • Survey for CSGF
  • Computing tool
  • Teaching with technology -- GTC
  • Effective Assessment -- GTC
  • ievobio?

Meetings, etc

  • Grad Teaching Community -- tonight
  • Open Science Meeting -- Weds