Calculating the inverse normal distribution

thor has asked for the wisdom of the Perl Monks concerning the following question:

Replies are listed 'Best First'.
Re: Calculating the inverse normal distribution by randyk (Parson) on Nov 22, 2005 at 22:44 UTC
The `ndtri` function of Math::Cephes can do this.	[reply]
Re: Calculating the inverse normal distribution by QM (Parson) on Nov 22, 2005 at 21:26 UTC
I'll preface this by saying I know just enough to be dangerous. But since you haven't got a reply yet, here goes... Do you want the distribution function? Doesn't the mean and standard deviation determine that uniquely? Why do you need the target percentage? I've seen the IGD page on Mathworld, and frankly I'd be concerned about coding that correctly, efficiently, etc. I note that Mathematica doesn't even have it yet. Perhaps you can clarify your question (or perhaps it's just me in over my head). -QM -- Quantum Mechanics: The dreams stuff is made of	[reply]
Re^2: Calculating the inverse normal distribution by thor (Priest) on Nov 22, 2005 at 21:40 UTC
Hmm...that's not quite it. What I have is the standard Gaussian distribution with a known mean (0) and standard deviation (1). Let's call that G(x). I also have a fixed percentage, let's say 0.85. What I need to find is a point z such that integral(G(x), -infinity, z) = 0.85. Does that make more sense? thor `Feel the white light, the light within Be your own disciple, fan the sparks of will For all of us waiting, your kingdom will come`	[reply]
Re: Calculating the inverse normal distribution by jeffguy (Sexton) on Nov 22, 2005 at 21:51 UTC
I googled: perl inverse gaussian distribution and followed the link trail to: http://home.online.no/~pjacklam/notes/invnorm/impl/acklam/perl/ltqnorm.pl Though I haven't looked at it in detail, it looks like it's what you want. Update: Wrapped it in a test script -- it's what you want.	[reply]
Re: Calculating the inverse normal distribution by swampyankee (Parson) on Nov 22, 2005 at 21:34 UTC
There is a `Statistics::Distributions` module; but I'm sure you know about that. I'm a bit confused by your post (I'm easily confused, so don't be offended), in that you're saying you've got the distribution, the mean, the standard deviation, and a target percentage. Isn't this over-specified? Or do you want to be able to get z given a target percentage and vice versa? Scanning CPAN does seem to indicate you'll have to roll your own; you might have success at http://lib.stat.cmu.edu, but you've probably already looked there. Update: One of thor's replies un-confused (or, at least, I think it did) me; `Statistics::Distributions` gives `n = F(z)`, but he needs the inverse of this. i.e., `z = F^-1(f)`. emc	[reply] [d/l] [select]
Re: Calculating the inverse normal distribution by aufflick (Deacon) on Nov 23, 2005 at 00:59 UTC
Everyone in this thread is WAY too smart for their own good ;)	[reply]
Re^2: Calculating the inverse normal distribution by spiritway (Vicar) on Nov 23, 2005 at 05:40 UTC
Including you?	[reply]
Re^3: Calculating the inverse normal distribution by aufflick (Deacon) on Nov 23, 2005 at 05:45 UTC
Sorry - circular reference there. Perhaps I should have implemented my comment with WeakRef :)	[reply]