Friday: wrightscape, ggplot, wp memory

more sub-model comparisons

wrightscape sub-models:

  • a1 independent alphas, global theta, sigmas

  • bm “brownie” (alpha = 0, indep sigmas)

  • a2 independent alpha, theta, global sigma

  • s1 indep sigma, global alpha, theta

  • s2 indep sigma, theta, global alpha

Basic Nelder Mead

Simulated annealing

associated parameter estimates

This approach is really not finding a very robust solution.

Plotting with stats_summary()

Using stats_summary instead of melt and cast computations and line and ribbon geometries for adding statistical layers to plots.  For instance, from plots.R in SDP code:


ggplot(data = subset(dat, variable == "unharvested"),
           aes(time, value, group = L1)) + geom_line(alpha=.2) +
  # shows the mean +/i mult * sd , requires Hmisc
  stat_summary(fun.data = mean_sdl, geom="smooth", mapping=aes(group = 1), mult=1)

Note that this needs thee aesthetic mapping (group=1) to know what to do. I’m not clear exactly why.

Note also that “mult” is an additional option, and is passed to the function, mean_sdl, specifying the multiplier in front of the standard deviation.

The documentation of possible function definitions in stat_summary aren’t entirely clear. This can take fun.y to apply the function across y values, for instance, to calculate just the mean:


ggplot(data = subset(dat, variable == "unharvested"),
       aes(time, value, group = L1)) + geom_line(alpha=.2) +
    stat_summary(fun.y = mean, geom="line", mapping=aes(group = 1))

Note that we’ve also changed the geometry to line, since we’re not returning ymin/ymax information.

We could add separate functions for the lower and upper bounds, fun.ymin and fun.ymax instead. Again we’d need a geometry that knew what to do with these – for instance, “smooth” or “errorbar”.

There’s quite a collection of convenient functions whose wrappers are provided through the package Hmisc, such as the mean_sdl choice shown here, which are illustrated well in the stat_summary examples.

Wordpress memory management

Wow, my saga continues for RStudio server running on Debian Lenny…