Yuriy Koblents-Mishke wrote:
For example, in my seventeen years of experience with economic modeling, we worked every day with normal, log-normal, Poisson, Pareto, and many other distributions, but I cannot remember a singe case when a single researcher used spherical harmonics or elliptic integrals. Our economists, probably, never hear these words, and most likely even mathematicians all but completely forgot them. I can imagine easily my former colleagues intimidated by special functions when browsing documentation, and putting aside the library completely. A typical user of special functions hardly will be intimidated by statistical distributions, but more often than not he will regard this part of manual as a mildly distracting noise.
Of course, gamma is gamma and beta is beta, but this message is about manual, not about codes.
Nod, understood. For many users you are completely correct of course, though for scientific programmers the distinction is less clear cut IMO. It might be possible to split the manual into two separate chapters (not sure about that though, might be a lot of work getting the links sorted out), but at the very least if the library is accepted we can include a direct link from the library index to the distribution-only TOC (see http://freespace.virgin.net/boost.regex/toolkit/html/math_toolkit/dist.html for example). There will always be a lot of cross-linking between the two - I don't believe it would be sensible to try and duplicate accuracy data in both places for example - the docs would quickly become impossible to maintain. There will also be a lot of curious folks who want to know how the distributions and the underlying special functions are implemented - either because they're just curious - or because they need to do a quality-audit on the code before recommending it to their workplace. On the other hand maybe I'm over-estimating how many folks fall into that category, just 'cos those are the kinds of people that get in touch.... hmmm, who knows.
And, by the way, Pareto is not mentioned in the manual.
That's one we haven't implemented yet, we've tried to implement a core set of distributions that were common to the majority of other statistics packages. We're slowly getting requests for more as various people find that their favorite isn't present. So basically, please be patient, but don't be shy in lobbying for for anything we've missed - and of course submissions are always welcome :-) John.