[rkward-devel] Distribution plugins (was Re: feedback)

[Thomas F. <tho...@ru...>]

Hi Aaron, and others,

On Sun, 23 Nov 2014 10:35:24 +0900
Aaron Batty <ab...@sf...> wrote:
> Piggybacking on Meik's feedback comments, I though I'd let you guys
> know what's going on in my undergraduate stats class, where we're
> using RKWard.

well, while touching on various areas of the code, I'm trying to work
on some of your feedback. Right now, I'm working to add i18n() to the
distribution plugins. And - as this will mean touching quite a bit of
that code - I'm wondering how to best address your comments regarding
those.

As a small bit of background information, IIRC, Stefan
originally modeled these dialogs on Rcmdrs example - many years ago.

> 3) Distributions dialogs
> 
> I only started using these this semester, so I hadn't noticed, but
> these are unintuitive enough that I decided to just get the students
> to do them in the R Console.
> 
> In the "r" dialogs (i.e., the ones that call rbinom and rgeom, etc.),
> the values that will be fed into R are filled in by default. Why?
> That seems to communicate, for example, that the "normal" thing to do
> when looking at a binomial probability is to set the number of
> "successes" to 0.95, and only 1 trial, and with a 50% probability,
> when actually, these are likely to all be different. I think these
> should just be blank.

You mean the p and q dialogs (pbinom, qbinom), right? Clearly for some
things, default values do not make too much sense. For "vector of
probabilities/quantiles", quite obviously so. For number of trials,
arguably, too - but that would mean it cannot be represented in a
spinbox (which simply does not have a "blank" value). For probability
of success in each trial, .5 sounds like a sensible enough default to
me, even if you are quite likely to adjust it.

So what are your thoughts on how to deal with this (also for all the
other distributions)?
 
> "Vector of quantiles" really ought to be renamed "Number of
> successes." I'll be honest here: I couldn't figure out what I was
> supposed to do here, so the only way I figured out how to use these
> dialogs was by clicking the "Code" button to see what commands it
> would be calling, and then reading the CRAN documentation on those
> functions. The CRAN documentation was much clearer.
>
> It would be nice if there was another calculator that called the "d"
> functions as well... Not that I've really ever used either of these,
> but they are pretty important for teaching undergraduate stats.

Well, what I have in mind so far is the following:
- Merge probabilities and quantiles plugins into one (which could also
  calculate density), with a <radio>-control on top to select the mode
  of operation (
    calculate:
     O probabilities of given quantiles
     O density at given quantiles
     O quantiles for given probabilities
  )
- For the discrete distributions, add a UI label below this, reading:
  "Note: Quantiles correspond to number of successes in 'size' trials in
  the context of the binomial distribution" (and appropriately for the
  other discrete distributions, or at least the ones I understand).
- Change the input of probabilities / quantiles to use a
  <matrix>-element (i.e. a column of spreadsheet cells)
- This will be empty by default.
- Change the output to be tabular, i.e. a table with columns
  for quantiles and probs (or densities), and rows for each input value.

Does that sound like a useful plan?

Regards
Thomas