Menu
▾
▴
#286 Submission: Beta dist and Incomplete Beta Function
Hi,
I've implemented the Beta distribution and also provided an implementation of the incomplete Beta function (which is the CDF for the Beta distribution).
I've tested the incomplete Beta function quite a bit (as shown), and think that it is now robust. However, I probably didn't make the most efficient implementation. An alternative would be the collection of algorithms in DCDFLIB (http://people.scs.fsu.edu/~burkardt/f_src/dcdflib/dcdflib.html), but it is quite involved -- if this implementation is sufficient, so much the better.
Eric
Logged In: YES
user_id=816411
Originator: YES
File Added: beta.zip
Logged In: YES
user_id=816411
Originator: YES
I'm attaching a version of the incomplete beta function that is significantly faster (about 10 faster for some values) than the original version I posted.
File Added: beta.zip
Logged In: YES
user_id=816411
Originator: YES
File Added: beta.zip
Logged In: YES
user_id=816411
Originator: YES
File Added: beta.zip
Logged In: YES
user_id=816411
Originator: YES
Attaching a new version. Fixes an edge case that can make the routine hang. Also introduces a short-cut in case of integer value for b.
File Added: beta.zip
Logged In: YES
user_id=816411
Originator: YES
File Added: beta.zip
Logged In: YES
user_id=816411
Originator: YES
File Added: beta.zip
Logged In: YES
user_id=816411
Originator: YES
File Added: beta.zip
Improved beta distribution
Logged In: YES
user_id=816411
Originator: YES
This is a better implementation for the beta distribution. First, I extended the useful range for the pdf. Second, I implemented a much faster algorithm for the incomplete beta function, using a continued fraction from Abramowitz & Stegun. (I used the GNU Scientific Library for inspiration, but ended up using a different continued fraction expansion than they used, so there's no GPL concern here. The speeds are comparable.)
File Added: beta.zip
Logged In: YES
user_id=400048
Originator: NO
Incorporated this. One thing to think about:
pdf-beta a b 0.0 or 1.0 are edge cases that should give either 0 or infinity (depending on a and b).
Right now, they give 0 only.