Category Archives: Phylogenetics

Monday: treebase manuscript writing

Catch­ing up on unfin­ished work: today has been spent mostly work­ing over my lit­tle tree­base appli­ca­tions note. I’m still decid­ing what/how to present in terms of the meta-analysis sec­tion — look­ing at the clas­sic trends on diver­si­fi­ca­tion rate sta­tis­tics: \(\gamma\) sta­tis­tic and diver­si­fi­ca­tion rate, and depen­dence on num­ber of taxa. Not sure how best to

Is your phylogeny informative?

Yes­ter­day my paper   appeared in early view in Evo­lu­tion (author’s preprint),1 so I’d like to take this chance to share the back-story and high­light my own view on some of our find­ings, and the asso­ci­ated pack­age on CRAN.2 I didn’t set out to write this paper.  I set out to write a very dif­fer­ent

wrightscape examples

  kt and open seem the best focal traits. Unclear what good null traits to use would be. Unclear if any­thing is gained by indep thetas in this exam­ple. Per­for­mance of alpha v theta? Edited plot­ting of like­li­hood com­par­isons, just grouped by trait to get com­mon axis. (Dif­fi­cult to get intel­li­gent zoom­ing on facet_grid to ignore the

Tuesday: pmc package, latex to word attempts

Back from Oz. Evo­lu­tion sub­mis­sion: Latex to word Final sub­mis­sion needs editable copy.  No great solu­tions for latex to word, despite quite a few options: tex2rtf is a fast solu­tion to get some­thing, but not suc­cess­ful on images or equa­tions.  Ended up going with this after all. pan­doc is a generic con­verter to many types, but has only

Thursday

Another day of work­ing on 4 dif­fer­ent projects at once. Warn­ing sig­nals review Alan pulses man­u­script review Paper dis­cussing the data you want \( \neq \) the data you want No way in Anchoveta data, look at time series: Exam­ple in the cod data using only up until 1991 data: Way for­ward — con­struct the step-wise pre­dic­tor of prob­a­bil­i­ties. 

Wednesday — wrightscape runs and various other things

PDG Con­trol plots for chang­ing costs Report out to Train­ing Prob­lem II group. ggplot note defin­ing func­tions that return a ggplot object can be tricky. If you com­pute some objects in the plot func­tion (stats etc), those stats are not stored by the object (due to lazy eval­u­a­tion, usu­ally a very nice time-saving fea­ture), so the plot

Tuesday — two insights

Con­trol & Opti­miza­tion Dominique Bon­vin visit two-step con­trol (opti­mize para­me­ters, opti­mize mean-square error between model and mea­sure­ment also by para­me­ter adjust­ment) fails when real­ity isn’t in the model set.  Prob­lem is under-determined (N free vari­ables, 4N equa­tions).  (con­verges, but onto the wrong opti­mum). Bet­ter solu­tion — adjust the penalty/regularization terms to to get bet­ter agree­ment between

More labrid plots

Intra­mandibu­lar like­li­hood com­par­isons Para­me­ter zooms: Pha­ryn­geal like­li­hood com­par­isons Both shifts:  

Release of constraint

      (uncer­tainty esti­ma­tion to follow).    

Friday — updated wrightscape results — simulated examples

Con­sider two sim­u­lated datasets, one under a release of con­straint (left) one under an increase in sigma (right).  Can we cor­rectly dis­crim­i­nate between these mod­els? Para­me­ter dif­fer­ences: Like­li­hood dif­fer­ences: Yup. Repeated with effect-size scaled to be roughly equiv­a­lent for both traits.  Won­der­ing if it’s eas­ier to plot para­me­ters out sep­a­rately like this.  Error bars show