GOSSET: A GENERAL-PURPOSE PROGRAM FOR DESIGNING EXPERIMENTS
N. J. A. Sloane and R. H. Hardin
Internet and Network Systems Research Center,
AT&T Shannon Labs, Florham Park, N.J. 07932-0971 U.S.A.
This is the home page for Gosset. It contains sections called
Last modified May 07, 2003.
This replaces the "README" file that was included
with older releases of Gosset.
Gosset is a flexible and powerful program for constructing experimental designs.
It runs under Unix, Linux and Mac OS X.
The following are some of its features.
-
Variables may be discrete or continuous (or both).
Discrete variables may be numeric or symbolic (or both).
Continuous variables may range over a cube or a ball (or both).
-
The variables may be required to satisfy linear equalities or inequalities.
-
The model to be fitted may be any low degree polynomial (e.g. a quadratic).
Much more general models can also be used - see Section 3.12
of the manual.
-
The number of observations is specified by the user.
-
The design may be required to include a specified set of points.
(so a sequence of designs can be found, each of which is optimal
given that the earlier measurements have been made).
-
The region where the model is to be fitted need not be the same as
the region where measurements are to be made
(so the designs can be used for extrapolation).
-
The following types of designs can be requested: I-, A-, D- or E-optimal,
the same but with protection against loss of one trial,
or packings (if no model is available).
-
Blocked designs and designs with correlated errors can also be obtained.
-
The algorithm is powerful enough to routinely minimize functions
of 1000 variables (e.g. can find optimal or nearly optimal designs
for a quadratic model involving 12 variables).
-
An extensive library of precomputed optimal designs
is included for linear and quadratic designs in the cube,
ball and simplex, involving up to 14 variables.
-
The user does not have to specify starting points for the search.
The user also has control over how much effort is expended by the algorithm,
and can if desired monitor the progress of the search.
-
Applications so far include VLSI production, conductivity of diamond films,
growth of protein crystals, flow through a catalytic converter,
laser welding, etc.
Gosset is described in more detail in the following documents.
-
R. H. Hardin and N. J. A. Sloane,
"Gosset: A General-Purpose Program for Constructing
Experimental Designs (Second Edition)"
[postscript file],
2003.
(Also included with the source files.)
[When viewing this postscript file with ghostview, set the
"PageSize" option, the 6th button at the top,
to "Letter" rather than
"Automatic" or "BBox"]
-
R. H. Hardin and N. J. A. Sloane, "A New Approach
to the Construction of Optimal Designs", J. Statistical
Planning and Inference, vol. 37, 1993, pp. 339-369
[Abstract,
postscript or
pdf].
-
R. H. Hardin and N. J. A. Sloane,
"Computer-Generated Minimal (and Larger) Response Surface Designs: (I) The Sphere"
(Abstract,
pdf,
ps).
-
R. H. Hardin and N. J. A. Sloane,
"Computer-Generated Minimal (and Larger) Response Surface Designs: (II) The Cube"
(Abstract,
pdf,
ps).
For a limited time you can download a beta-test version of Gosset at no charge.
We are doing this for several reasons:
- to collect feedback from users
- to collect information about applications
- to encourage the use of Design of Experiments in industry
Of course, "anything free comes with no guarantee".
If you download the program, please send email to Neil Sloane
at this address:
njas@research.att.com,
giving your name, affiliation, and postal address, for our records.
If you find it useful in some real application, or teaching a course, we would
like to get some comments. For example, a real-life example of
how you used it (of course, change any proprietary data).
If you publish anything that makes use of a design that was partly obtained using
Gosset, please cite Gosset, just as you would acknowledge a design
that a colleague constructed for you.
The citation could read something like this:
R. H. Hardin and N. J. A. Sloane,
GOSSET: A general-purpose program for designing experiments,
www.research.att.com/~njas/gosset/.
Also we would appreciate it if you could inform us about such a
paper.
The files can be obtained from the ftp site
ftp://ftp.research.att.com/
You need to obtain two files:
index.html (this file) and
codemart.cpio (about 5 megabytes)
Go to directory "dist", then "njas", then "gosset",
where you will find the files.
Warning: it seems to take some browsers a long time to
find this site!
Memo to colleagues: if you are inside the Labs firewall,
you need to set
"Use Passive FTP (for firewall and DSL modem compatibility)"
This setting is under Tools / Internet Options / Advanced
The following instructions are an updated version of those in Section 9 of the manual.
System administrators please note: if you are installing
gosset for multiple users, the instructions are slightly different
- see below.
INSTALLING A PRIVATE COPY OF GOSSET
- First choose a base directory for gosset.
Subdirectories will be created off this for work on individual problems.
- Place the files index.html (this file) and
codemart.cpio in the base directory.
- Change directory to the base directory.
To print the manual, which is about 130 pages long, type
$ lp manual.ps
(possibly replacing "lp" by your local printer command, e.g. "lpr").
(If you view this postscript file with ghostview, set the
"PageSize" option, the 6th button at the top,
to "Letter" rather than
"Automatic" or "BBox")
- Extract everything in codemart.cpio :
$ cpio -icu < codemart.cpio
(some users have found that it is necessary to use
$ cpio -iu < codemart.cpio
instead)
and compile several files:
$ cc -w M*.c -o gosset -lm
$ rm */vtrace */vvv */moments */interp
(The latter command is needed only if you already
had a copy of gosset. It removes some out-of-date files.)
The archive file codelib.a does not need to be touched.
- To find a design, start by entering gosset :
$ gosset
The program responds by asking you to name a working subdirectory:
please type 'cd something' to name a local directory for your work
At this point you will probably enter the specifications
for a new design - see Sects. 2, 3, .... of the manual.
- First choose a directory for the gosset source files, say
/u/g/sources.
- Place index.html and codemart.cpio in /u/g/sources
- Change directory to /u/g/sources.
To print the manual, which is about 130 pages long, type
$ lp manual.ps
(possibly replacing "lp" by your local printer command, e.g. "lpr").
(If you view this postscript file with ghostview, set the
"PageSize" option, the 6th button at the top,
to "Letter" rather than
"Automatic" or "BBox")
- Extract everything in codemart.cpio :
$ cpio -icu < codemart.cpio
(some users have found that it is necessary to use
$ cpio -iu < codemart.cpio
instead)
and compile several files:
$ cc -DGOSSETSRC=\"/u/g/sources\" -DCODELIB=\"/u/g/sources\" M*.c -o gosset -lm
which will create a version of gosset that expects the source files to be in
/u/g/sources.
The escapes are necessary on the quotes, because they are part of
the definition.
- The executable file gosset can be moved anywhere, say into
/usr/local/bin , and can be executed from anywhere.
- Be warned, however - working directories will be created by
gosset wherever the user is, instead of from a single base
directory. So the user must exercise some discipline when
invoking gosset, to avoid chaos in the directory structure.
- If gosset was already installed on your system, users should
execute
$ rm vtrace vvv moments interp
in each working subdirectory, to remove out-of-date files.
- The archive file codelib.a does not need to be touched.
- When someone wants to find a design, they change to a base
directory (typically $HOME/gosset), from which working subdirectories
will be created, and enter gosset:
$ gosset
The program responds by asking them to name a working subdirectory:
please type 'cd something' to name a local directory for your work
At this point they will probably enter the specifications
for a new design - see Sects. 2, 3, .... of the manual.
There have been three kinds of applications:
Applications to industrial problems
(It is inevitable that the most spectacular applications
do not get published!)
-
J. Hagen, M. Heinen and W. Salber:
DOE in der Optimierung von Ottomotoren mit vollvariabler Ventilsteuerung;
paper for the conference "Design of Experiments in der Motorenentwicklung",
Haus der Technik, Dec. 2001.
-
Charles W. Carter has used Gosset successfully in designing several
experiments.
- Optimization of Resist Formulation and Processing with
Disulfone Photo Acid Generators Using Design of Experiments,
O. Nalamasu, A. Freeny, E. Reichmanis, N.J.A. Sloane and L. F. Thompson,
AT&T Bell Labs Memorandum, 1993.
Applications to theoretical understanding of optimal designs
-
R. H. Hardin and N. J. A. Sloane,
"Computer-Generated Minimal (and Larger) Response Surface Designs: (I) The Sphere"
(Abstract,
pdf,
ps).
-
R. H. Hardin and N. J. A. Sloane,
"Computer-Generated Minimal (and Larger) Response Surface Designs: (II) The Cube"
(Abstract,
pdf,
ps).
-
Several other papers are in preparation.
Applications to mathematical problems
- Packing Lines, Planes, etc.: Packings in Grassmannian Space
[Abstract,
postscript, pdf],
J. H. Conway, R. H. Hardin and N. J. A. Sloane,
Experimental Math., 5 (1996), pp. 139-159.
- Codes (Spherical) and Designs (Experimental),
R. H. Hardin and N. J. A. Sloane,
Different Aspects of Coding Theory, ed. A. R. Calderbank, AMS Series Proceedings Symposia Applied Math., Vol. 50,
1995, pp. 179-206.
- McLaren's Improved Snub Cube and Other New Spherical Designs in Three Dimensions
[Abstract,
postscript, pdf]
[note:
Fig. 1a and
Fig. 1b
are in separate files]
R. H. Hardin and N. J. A. Sloane,
Discrete and Computational Geometry,
15 (1996),
pp. 429-441.
- Minimal-Energy Clusters of Hard Spheres
[Abstract,
postscript, pdf]
[note:
Fig. 11 ,
Fig. 12
and "photos" of the putatively optimal clusters of
4 to 10 and
13 to 20
spheres are in separate files],
by N. J. A. Sloane, R. H. Hardin, T. S. Duff and J. H. Conway,
Discrete Computational Geom.,
14 (1995),
pp. 237-259.
- Expressing
(a^2 + b^2 + c^2 + d^2 )^2
as a Sum of 23 Sixth Powers
[Abstract,
postscript, pdf],
R. H. Hardin and N. J. A. Sloane,
Journal of Combinatorial Theory, Series A,
68 (1994),
pp. 481-485.
Also DIMACS Technical Report 93-54, August 1993.
- New Spherical 4-Designs
[ Abstract,
postscript, pdf],
R. H. Hardin and N. J. A. Sloane,
Discrete Mathematics,
106/107 (1992),
pp. 255-264,
(Topics in Discrete Mathematics, vol. 7, "A Collection of Contributions in Honour of Jack Van Lint", ed. P. J. Cameron and H. C. A. van Tilborg, North-Holland,
1992.).
- Spherical Designs in Four Dimensions (Extended Abstract)
[Abstract, pdf, ps],
N. J. A. Sloane, R. H. Hardin and P. Cara,
Proceedings Information Theory Workshop (Paris, April 2003),
2003.
For further information contact
N. J. A. Sloane
Information Sciences Research Center
Room C-233
AT&T Shannon Labs
180 Park Avenue
Florham Park NJ 07932-0971 USA
Email address: njas@research.att.com
fax: 973 360 8178
Gosset was developed at AT&T during 1991-2003 by
R. H. Hardin and N. J. A. Sloane, and is copyright 1991-2003
by R. H. Hardin and N. J. A. Sloane.

See also:
Neil Sloane's home page