PowerStation/Numerical Recipes questions
Author 
Message 
Troy Alton Ada #1 / 11

PowerStation/Numerical Recipes questions
Hi, folks. I'm a grad student who needs to do a lot of mathematical processing on hundreds of different functions for my dissertation, and I've looked at using Gauss, IMSL, and the procedures in Numerical Recipes. My fortran is slightly rusty (I've been using <shudder> QB and VB for a couple of years), but using Gauss would basically entail learning another language. I've got access to a VAX copy of IMSL, but I'd like to keep all my programming on my PC, so I'm seriously thinking about just using the Numerical Recipes procedures and a PC version of Fortran. I saw PowerStation at the book store the other day and noticed that it supposedly has the Numerical Recipes algorithms already coded in for you (did I understand that right?), so I'm thinking of buying it. But before I do, there's a couple of things I was wondering about: 1) If anyone out there has used the Numerical Recipes algorithms, is the subroutine library complete (i.e., do they provide all of the routines that are in the books of the same name)? 2) I've currently got Fortran 5.0. Would there be any speed advantage for going with the PowerStation (I've got a 486/DX 266 with 12 MB of RAM, so configuration is not a problem)? 3) I looked through the UseNet Fortran list, and I noticed a lot of messages about bugs in PowerStation. Any comments, anyone? (I will be doing math and string processing  no graphics, etc...  so I'm especially concerned about any problems of that nature). 4) Any comments in general on the downhill simplex minimzation program Amoeba in Numerical Recipes? I plan on using it for Maximum Likelihood estimation, and would appreciate hearing from anyone else who's used it. Thanks. Troy Adair

Fri, 29 Aug 1997 17:12:24 GMT 


Petros Dafniot #2 / 11

PowerStation/Numerical Recipes questions
Quote: [... comments deleted ...] Quote: >1) If anyone out there has used the Numerical Recipes > algorithms, is the subroutine library complete (i.e., do they > provide all of the routines that are in the books of the same name)?
Yes. There is a "smart" extractor that extracts only the routines you want. BTW, you do get sources of the NUMERICAL RECIPES not just an object library. (in fact you will need to build the library if you need it) Quote: >2) I've currently got Fortran 5.0. Would there be any speed > advantage for going with the PowerStation (I've got a 486/DX > 266 with 12 MB of RAM, so configuration is not a problem)?
Probably, you do not comprehend entirely the nature of the problem. MSFortran v5.0 is a dead product (it died at v5.1); in addition, it is a DOS compiler => 640Kb limit (ok you can get a bit higher with overlays and a lot higher with Windows and v5.1) but you still have segmented memory to deal with. Powerstation, OTOH, is a 32bit flat memory compiler that comes with one of the best DOS extenders a mini version of PharLap's. If you will never have to deal with big arrays/data, etc then stick with v5.0; else, you will need to use the Powerstation. Another significant edge that the Powerstation has is the number of DOS functions available: e.g. change directory, get list of files matching a particular pattern, graphics, etc that are not all available under MS Fortran v5.x Quote: >3) I looked through the UseNet Fortran list, and I noticed a lot > of messages about bugs in PowerStation. Any comments, anyone? > (I will be doing math and string processing  no graphics, etc... >  so I'm especially concerned about any problems of that nature).
Microsoft is an easy target for many of us (us = PC fortran users) because it does not respond fast to problems/bugs. However, you need to realize that MS Powerstation due to its price (~$85 academic price) has sold thousands(?) of copies and so much more people get into the bugs finding story. All compilers have bugs. That does not mean that MS Powerstation is so bad; it is just not the best Fortran PC compiler (I believe Lahey and Salford have the two best but quite pricy compilers). Quote: >4) Any comments in general on the downhill simplex minimzation program > Amoeba in Numerical Recipes? I plan on using it for Maximum > Likelihood estimation, and would appreciate hearing from anyone else > who's used it.
Sorry. Can't help you on that... Quote: Petros Dafniotis
ps: All in all, if I were you I would stick with the VAX and the DEC Fortran compiler. There is no other Fortran compiler as good as that one and the VMS de{*filter*} is the best in the world, hands down!

Fri, 29 Aug 1997 20:41:03 GMT 


Dan P #3 / 11

PowerStation/Numerical Recipes questions
Quote: >as that one and the VMS de{*filter*} is the best in the world, hands >down!
Ever tried dde on a Domain/OS box? Dan  Dan Pop CERN, CN Division
Mail: CERN  PPE, Bat. 31 R004, CH1211 Geneve 23, Switzerland

Fri, 29 Aug 1997 22:51:11 GMT 


che.. #4 / 11

PowerStation/Numerical Recipes questions
Quote: >Hi, folks. > I'm a grad student who needs to do a lot of mathematical >processing on hundreds of different functions for my dissertation, >and I've looked at using Gauss, IMSL, and the procedures in >Numerical Recipes. My Fortran is slightly rusty (I've been using ><shudder> QB and VB for a couple of years), but using Gauss would >basically entail learning another language. I've got access to a >VAX copy of IMSL, but I'd like to keep all my programming on my PC, >so I'm seriously thinking about just using the Numerical Recipes >procedures and a PC version of Fortran. I saw PowerStation at >the book store the other day and noticed that it supposedly has >the Numerical Recipes algorithms already coded in for you (did I >understand that right?), so I'm thinking of buying it. But before >I do, there's a couple of things I was wondering about: >1) If anyone out there has used the Numerical Recipes > algorithms, is the subroutine library complete (i.e., do they > provide all of the routines that are in the books of the same name)?
Yes, every routine in the book is on the disk. Access is made incredibly easy with a menudriven help file and automatic decompression of files. Quote: >2) I've currently got Fortran 5.0. Would there be any speed > advantage for going with the PowerStation (I've got a 486/DX > 266 with 12 MB of RAM, so configuration is not a problem)?
I believe you will get a significant speed improvement, but how significant I don't know. Anyone else know? Quote: >3) I looked through the UseNet Fortran list, and I noticed a lot > of messages about bugs in PowerStation. Any comments, anyone? > (I will be doing math and string processing  no graphics, etc... >  so I'm especially concerned about any problems of that nature).
The first version was quite buggy. The "update" (still version 1.0) is much better. I do only numerical stuff and haven't found any bugs since I installed the update (which was free from MS and should be the currently sold program by now). Quote: >4) Any comments in general on the downhill simplex minimzation program > Amoeba in Numerical Recipes? I plan on using it for Maximum > Likelihood estimation, and would appreciate hearing from anyone else > who's used it.
Coincidently, I plan on using this soon, but have not yet. Quote:

Sat, 30 Aug 1997 08:18:03 GMT 


Brian G McHenry  McHenry Consultan #5 / 11

PowerStation/Numerical Recipes questions
: Hi, folks. : I'm a grad student who needs to do a lot of mathematical : processing on hundreds of different functions for my dissertation, : and I've looked at using Gauss, IMSL, and the procedures in : Numerical Recipes. My Fortran is slightly rusty (I've been using : <shudder> QB and VB for a couple of years), but using Gauss would : basically entail learning another language. I've got access to a : VAX copy of IMSL, but I'd like to keep all my programming on my PC, : so I'm seriously thinking about just using the Numerical Recipes : procedures and a PC version of Fortran. I saw PowerStation at : the book store the other day and noticed that it supposedly has : the Numerical Recipes algorithms already coded in for you (did I : understand that right?), so I'm thinking of buying it. But before : I do, there's a couple of things I was wondering about: : 1) If anyone out there has used the Numerical Recipes : algorithms, is the subroutine library complete (i.e., do they : provide all of the routines that are in the books of the same name)? : 2) I've currently got Fortran 5.0. Would there be any speed : advantage for going with the PowerStation (I've got a 486/DX : 266 with 12 MB of RAM, so configuration is not a problem)? : 3) I looked through the UseNet Fortran list, and I noticed a lot : of messages about bugs in PowerStation. Any comments, anyone? : (I will be doing math and string processing  no graphics, etc... :  so I'm especially concerned about any problems of that nature). : 4) Any comments in general on the downhill simplex minimzation program : Amoeba in Numerical Recipes? I plan on using it for Maximum : Likelihood estimation, and would appreciate hearing from anyone else : who's used it. : Thanks. : Troy Adair

Sat, 30 Aug 1997 13:47:48 GMT 


Chris Wilki #6 / 11

PowerStation/Numerical Recipes questions
Quote: >4) Any comments in general on the downhill simplex minimzation program > Amoeba in Numerical Recipes? I plan on using it for Maximum > Likelihood estimation, and would appreciate hearing from anyone else > who's used it.
This is a very slow style of algorithm only to be used in dire circumstances. Unless you've got a very good reason, you should try to use BFGS or conjugate gradient techniques for most maximisation problems. In fact, Maximum Likelihood problems have much more structure than generic maximistation problems (a bit like least squares problems) and I think you can often use a modified LevenbergMarquandt style method (approximating the second derivatives of the ML function using derivatives of the residuals). Unless you are fairly clear about what you are doing, you may find that learning the few commands you need in Gauss will cause you less grief in the long run than using inappropriate routines in Fortran. Chris

Sat, 30 Aug 1997 08:53:03 GMT 


Brian G McHenry  McHenry Consultan #7 / 11

PowerStation/Numerical Recipes questions
last post fubar'd, try num 2:
: Hi, folks. <stuff deleted> : 1) If anyone out there has used the Numerical Recipes : provide all of the routines that are in the books of the same name)? I believe the disk provided with Powerstation is a subset of the full disk you get direct. : 2) I've currently got Fortran 5.0. Would there be any speed : advantage for going with the PowerStation (I've got a 486/DX : 266 with 12 MB of RAM, so configuration is not a problem)? you will realize a SIGNIFICANT speed improvement (32 bit v 16 bit) The full glory of the speed is in numerical processing. plus the windows interface/programming environment is a plus. Of course, you can't create windows programs ( except NT ) : 3) I looked through the UseNet Fortran list, and I noticed a lot : of messages about bugs in PowerStation. Any comments, anyone? : (I will be doing math and string processing  no graphics, etc... :  so I'm especially concerned about any problems of that nature). There's a maintenance release (1.0a?) that I'd assume they now ship with the product. The bugs had to do with Linker incompatablilities misc other items. Of couse Dos hosting Windows hosting DosXtended executables leads to occasional fubar's. Make frequent backups of your Make files for they can get corrupted if one of your programs GPF's windows. I'd also recommend Win 3.11 as host. I can't recall a GPF since win 3.11 w/the maintenace release so either they've fixed it or I'm writing less flakey code. I guess the former! : 4) Any comments in general on the downhill simplex minimzation program : Amoeba in Numerical Recipes? I plan on using it for Maximum : Likelihood estimation, and would appreciate hearing from anyone else : who's used it. I've been playing with Amoeba and find it to be very interesting. however, depending on the function(s) you're messing with and the number of unknowns, it may not be the most appropriate. If you tie in a graphical depiction of the search and hunt procedure you'll find Amoeba can frequently go either down the wrong path ( if there are many) and/or get stuck on a local plateau. You'll find the documentation of the inputs to Amoeba point you back to Nelder & Mead, Yarbro, and/or Jacoby. I've obtained the documents and I'd suggest you review them prior to delving into the routine. You'll find them strewn about your school library. The NR folks changed the routine somewhat from the 1st release of the NR collection. The factors for reflection, contraction & expansion of the Amoeba are built in as the alpha, beta and gamma values I believe in the orig and in Nelder & Mead there was discussion of some question as to the appropriate size for these. NR has adopted the values suggested in Jacoby. (1., .5, 2. ) I'm guessing cause I haven't revisited the routine for some time Well that's about it, time I stopped procrastinating and got back to whatever it was I was doing. : Troy Adair
brian

Sat, 30 Aug 1997 21:50:16 GMT 


Jiangang Li #8 / 11

PowerStation/Numerical Recipes questions
Here is my experience with Powerstation and Numerical Recipes. 1. Powerstation with academic discount is great value. The numerical recipes alone would cost $50 dollars. No bug was found that affects computation. 2. If you need to spend more than half year on your code for thesis and with a lot of matrix, I would recommend Lahey Fortran 90 for $595. It is much more expensive but will cut your programming effort by at least half. I did a Bayes model for my PH.D. thesis and this is my reflection with my one month experience in Fortran 90. 3. The subroutine Ameoba can only do simple optimization. For more complicated ones you are advised to use two routines that need first order derivative. Among the two, frprmn seems a bit more stable than dfpmin. J.G. Liao

Sat, 30 Aug 1997 22:49:59 GMT 


Jorg La #9 / 11

PowerStation/Numerical Recipes questions
: Hi, folks. : I'm a grad student who needs to do a lot of mathematical : processing on hundreds of different functions for my dissertation, : and I've looked at using Gauss, IMSL, and the procedures in : Numerical Recipes. My Fortran is slightly rusty (I've been using : <shudder> QB and VB for a couple of years), but using Gauss would : basically entail learning another language. I've got access to a : VAX copy of IMSL, but I'd like to keep all my programming on my PC, : so I'm seriously thinking about just using the Numerical Recipes : procedures and a PC version of Fortran. I saw PowerStation at : the book store the other day and noticed that it supposedly has : the Numerical Recipes algorithms already coded in for you (did I : understand that right?), so I'm thinking of buying it. But before : I do, there's a couple of things I was wondering about: : 1) If anyone out there has used the Numerical Recipes : algorithms, is the subroutine library complete (i.e., do they : provide all of the routines that are in the books of the same name)? : 2) I've currently got Fortran 5.0. Would there be any speed : advantage for going with the PowerStation (I've got a 486/DX : 266 with 12 MB of RAM, so configuration is not a problem)? : 3) I looked through the UseNet Fortran list, and I noticed a lot : of messages about bugs in PowerStation. Any comments, anyone? : (I will be doing math and string processing  no graphics, etc... :  so I'm especially concerned about any problems of that nature). : 4) Any comments in general on the downhill simplex minimzation program : Amoeba in Numerical Recipes? I plan on using it for Maximum : Likelihood estimation, and would appreciate hearing from anyone else : who's used it. : Thanks. : Troy Adair
Hi Troy I cannot help you on V5.0 vs. Powerstation, I have used V5.1 till now and has just bought Powerstation but had no time yet to get my teeth into it! However, while doing my Ph D I used the NelderMead algorithm (which is the simplex, or hillclimbing, algorithm used in Amoeba) to adapt Kalman filters. It seemed to work well, the nelderMead algorithm is actually an improvement on the original simplex by Spendley, et al. Unfortunately it still suffers from some of the problems, e.g. converging on "false" optimum peaks. I suspect you will experience similar problems when applying it to maximum likelihoods. Let me know how this application is progressing! Regards

Tue, 02 Sep 1997 16:51:49 GMT 


Tom Ho #10 / 11

PowerStation/Numerical Recipes questions
Quote:
>Subject: Re: PowerStation/Numerical Recipes questions
>Date: 14 Mar 1995 00:18:03 GMT
deletions Quote: >2) I've currently got Fortran 5.0. Would there be any speed > advantage for going with the PowerStation (I've got a 486/DX > 266 with 12 MB of RAM, so configuration is not a problem)? >I believe you will get a significant speed improvement, but >how significant I don't know. Anyone else know?
It depends on the code, but you can expect a program using Powerstation to run in about 60% of the time it takes to run using MS Fortran 5.0. A rough guide is that with Powerstation you are using 32bit processing rather than 16bit. So the speedup will be more or less by a factor of two minus some reduction for the inefficiency of the system. However, for me, the main advantage of Powerstation over previous versions is the superb Windowsbased development environment. I even use it to develop code to run on UNIX boxes. Tom

Sat, 20 Sep 1997 03:00:00 GMT 


Julie Greenwoo #11 / 11

PowerStation/Numerical Recipes questions
Quote:
>However, for me, the main advantage of Powerstation > over previous versions is the superb Windowsbased development > environment. I even use it to develop code to run on UNIX boxes.
I like the environment too, but I wouldn't call it superb. I sorely miss the ability to tailor the editor the way I can in Programmer's Workbench. And how do I search through all project files? And why aren't my inline comments recognized as comments (as far as the text coloring is concerned)? :^) julie
www  http://www.oceanweather.com/~oceanwx/oceanwx.htm

Sat, 20 Sep 1997 03:00:00 GMT 


