home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.math.symbolic
- Path: sparky!uunet!gumby!wupost!spool.mu.edu!umn.edu!stecher
- From: stecher@atlas.socsci.umn.edu (Jack Stecher)
- Subject: Compiled Functions in Mma
- Message-ID: <C0nFt4.F0B@news2.cis.umn.edu>
- Sender: stecher@atlas.socsci.umn.edu
- Nntp-Posting-Host: daphne.socsci.umn.edu
- Organization: Department of Economics, Univ of Minn
- Date: Sun, 10 Jan 1993 17:41:27 GMT
- Lines: 29
-
- I've been trying to use compiled functions in Mma 2.1, and have run into
- some difficulties.
-
- First, the difficulty I suspect there isn't a way around: Does anyone know
- if there's a way to use Compile when the arguments are matrices? Didn't
- think so....
-
- Now, for one that just confuses me: I'm using Mma on a Dec/Ultrix type
- station, and don't know where to find the stats routines. It's easy enough
- to write the formula for a normal density, and the normal cdf can be written
- using Erf. I used Newton's method to solve iteratively for a normal quantile
- corresponding to a given probability.
-
- When I used uncompiled code for my everything, solving the normal quantile
- for .0001 took something like 8 seconds; with compiled code, it took something
- like half a second. But then I loaded a data set to do statistical calcula-
- tions on. The calculations used my compiled code within the routines (I
- was computing log-likelihood functions), and were incredibly slow. Well, after
- I finished, I timed the normal quantile for .0001, and it took something like
- 20 seconds, using compiled code.
-
- Anyone have any clues about where I should look for the problem?
-
- Jack Stecher
- stecher@atlas.socsci.umn.edu
- --
- A government big enough to give you all you want
- is big enough to take it all away.
- --Barry Goldwater
-