[R--gR] CoCo 1.6.1
Jens Henrik Badsberg
JensHenrik.Badsberg at agrsci.dk
Wed Jun 2 12:36:53 CEST 2004
Dear gR,
A new version, 1.6.1, of the CoCo bundle with CoCoCg is now available from
www.jbs.agrsci.dk/Biometry/Software-Datasets/CoCo/CoCo.1.6 and from
http://www.math.auc.dk/gr/material/CoCo/ .
For windows users: Remember to have cygwin1.dll somewhere in your path,
(e.g. in .../rw1090/bin if you do not have it in a more permanent place).
The CoCoCg package is now "more useful":
- CoCoCg and MIM produces the exact (to four digits after the decimal point)
same deviance (and DF) of all the 760 MIM-models on the Rats-data
(see table 4.4 of Introduction to Graphical Modelling, David Edwards,
or data(Rats) after library(CoCoCg) ).
- All the 1101 hierarchical interaction models (as defined in section 6.4
of Graphical Models, Lauritzen 1996) on Rats can be fitted by CoCoCg,
and 2 times 1101 tests all gives positive difference between deviances.
[ The two above: An other convergence criterion (difference between observed
and expected sufficient statistics) and full line search is now implemented,
and fix of updating single quadratic generators in CoCoCg. ]
- The degrees of freedom is also computed correct when specifying the
quadratic generators independent of the quadratic pair generators by the
"^" as in e.g. enterModel("[[ab]] / [[aby][abx]] / [[y^][abxy][x^]];")
[ Fix of term for intersection of quadratic generators in computing DF. ]
A package called CoCoGraph is now made by taking a part out of CoCoObject -
a) to be able to load CoCo and CoCoCg without dynamicGraph (loading dynamic-
graph does take some time with the current versions of R and dynamicGraph),
b) to highlight the interface to dynmaicGraph.
The four functions propertySet, propertyModel, returnVertexOrder and
returnJunctionTree can now be used without "data" in a more flexible
form, as, e.g.
str(returnJunctionTree(model = "IJ,IZ,ZY", continuous = "YZ"))
for the junction tree of graph 6.1 (c), page 176 in Lauritzen 1996.
Regards,
Jens Henrik Badsberg
More information about the R-sig-gR
mailing list