v091nm4y@ubvmsd.cc.buffalo.edu (Samir Haque) (10/04/90)
I have a copy of McClelland and Rumelhart's programs
Explorations in PDP but, alas not enough documentation.
Could someone please direct me to the references, or e-mail some
documentation on the use/limitations of these programs.
The exe files are : cl, ia, utils, src, iac, cs, pa, bp, aa.
I realize the books are available from the MIT Press, but don't know the
titles, this would be most helpful.
TIA,
Samir.
============================================================================
"There are things on Heavan and Earth, Horatio, Man was not meant to know."
-Hamlet (Minsky??) for499@uxh.cso.uiuc.edu (10/05/90)
The title is Explorations in Parallel Distributed Processing, ISBN
0-262-63113-X (pbk.). This is the book that comes with the programs
you mentioned. The companion books are Parallel Distributed Processing
Vol. I & II. (ISBN 0-262-63112-1 for the set, pbk.). Hope this information
is helpful.
B. T. GuanPLai@cup.portal.com (Patrick L Faith) (10/08/90)
I've been thinking of getting the pdp programs/workbook ... has anyone worked with it enough to say the level at which the projects are written ... i.e. is are these simple work book type programs or are they something that could do something useful ? PLai
eesnyder@boulder.Colorado.EDU (Eric E. Snyder) (10/09/90)
In article <34624@cup.portal.com> PLai@cup.portal.com (Patrick L Faith) writes: >I've been thinking of getting the pdp programs/workbook ... has anyone >worked with it enough to say the level at which the projects are written ... >i.e. is are these simple work book type programs or are they something >that could do something useful ? I have been working with these programs recently. I am by no means and AI expert but for my needs they are quite sufficient. I posed a question to this group a few weeks ago. With a few hours of serious study, I was able to adapt the back propagation program to solve the problem and give some meaningful results. To this extent, the programs are useful in the real world. On a related note: The back propagation program was running a little slow on my PC. Fortunately, the source code is included in the package and I was able to get the programs running on our departmental mainframe in about 10 min, drastically reducing run times. --------------------------------------------------------------------------- TTGATTGCTAAACACTGGGCGGCGAATCAGGGTTGGGATCTGAACAAAGACGGTCAGATTCAGTTCGTACTGCTG Eric E. Snyder Department of MCD Biology We are not suspicious enough University of Colorado, Boulder of words, and calamity strikes. Boulder, Colorado 80309-0347 LeuIleAlaLysHisTrpAlaAlaAsnGlnGlyTrpAspLeuAsnLysAspGlyGlnIleGlnPheValLeuLeu ---------------------------------------------------------------------------
pratt@yoko.rutgers.edu (Lorien Y. Pratt) (10/09/90)
>I've been thinking of getting the pdp programs/workbook ... has anyone >worked with it enough to say the level at which the projects are written ... >i.e. is are these simple work book type programs or are they something >that could do something useful ? I've used the bp program for two years on several large applied and research projects. For the money, they are an incredible deal compared with the simulators you can buy from people by neuralworks, etc, which cost in the thousands of dollars. I hit the limit of this package when I was running a network with >1 million training patterns, >5000 network weights, on a parallel machine without an optimizing C compiler (on the order of 10^11 arithmetic operations per network training session). The program ran, but excruciatingly slowly. I had to switch to a special-purpose Fortran simulator. The problem was not in the code's usability, but in its speed with the sun C compiler that I had available, which wasn't written with supercomputer-sized applications in mind. But, until this point, I was very happy with this package and strongly recommend it to anybody getting started with neural networks. I should also say that I've modified the PDP code extensively and found it to be modularly written and very easy to change reliably, despite my initial fears, based on the lack of in-line commentary in the code. A little time invested in reverse-engineering the package and working by analogy to other program modules allowed me to add new commands and change existing functionality with relative ease. -- ------------------------------------------------------------------- L. Y. Pratt Computer Science Department pratt@paul.rutgers.edu Rutgers University Hill Center (201) 932-4634 (Hill Center office) New Brunswick, NJ 08903, USA (201) 846-4766 (home)