Tuesday: PMC, confidence intervals; screen code-tricks

PMC

  • Final read through of manuscript from me and Peter.

  • Looking at confidence intervals in fitContinuous?  Package used to use the hessian from optim, but it was later removed due to errors.  I added this back manually and attempt a few estimates.  Given the Hessian (likelihood surface curvature) this is approximated by

    sqrt(diag(solve(out$hessian)))

  • See conf_interval_ex.R.  i.e for Geospiza tree is:

    [1] SE sigma: +/ 0.6763731 SE lambda: +/- 1.0338531 [2] boot lambda: (9.9e-08, 1) [3] boot beta: (0.0072 0.0591)

More crazy on large tree:  the bootstrap estimates are nice and narrow, while the Hessian estimate for lambda is ridiculous:

SE sigma: +/- 0.05773465  SE lambda: +/- 73.05777456
boot lambda: (0.452340436725362, 0.694239348471331)
boot beta: (0.143594988736606, 0.214849236562022)
  • added revolution-r packages to zero, parallelizes fitContinuous on my machine, but interestingly crashes it on zero when run on the large tree.  Hmm…

database packages

  • Scott’s updates to dryad.  Scott’s comment’s on workflow added to wiki.  Need to check over and test.

  • Fixes from Emmanuel to read.nexus.data in the ape package, testing in TreeBASE package.   Basic examples that return the alignments now implemented.   A few issues remain: need to track the metadata (but can’t tag this onto the list), and lots of character matrices don’t meet read.nexus.data’s requirements.

  • Discussion with Dave on his work.  Let a stochastic model parameter (binomial probability) evolve or not evolve on a tree.  Partition the uncertainty.  Considering some rather strange/arbitrary models.

Code tricks

Copy-paste in screen mode, very useful on remote machine:

Ctrl+a [

starts copy mode.  Navigate with vi-keys (arrows, etc), Hit spacebar to begin copy selection, spacebar again to end.  Paste with

 Ctrl+a ]

<span style="font-family: constantia, 'hoefler text', 'palatino linotype', serif; font-size: 16px; line-height: 24px; white-space: normal;">Note that git can of course always pull directly from the server, rather than pushing through github.     i.e. </span>

git remote add zero cboettig@ten.ucdavis.edu:/home/cboettig/Computing/RMendeley

git pull zero master

Perhaps worth configuring to share access to individual manuscript files.