Valued nodes #13
Valued nodes #13
Conversation
| ## Use the R_HOME indirection to support installations of multiple R version | ||
| PKG_LIBS = `$(R_HOME)/bin/Rscript -e "Rcpp:::LdFlags()"` | ||
| PKG_CPPFLAGS=-I../inst/include | ||
| PKG_CXXFLAGS += -Wno-sign-compare |
There was a problem hiding this comment.
removes annoying flags when building
There was a problem hiding this comment.
added some timing flags.
ifellows
left a comment
ifellows
left a comment
There was a problem hiding this comment.
Great work. A couple of comments.
| * Adds a statistic the absolute value of the difference between the values | ||
| * of a continuous nodal covariate | ||
| */ | ||
| template<class Engine> |
There was a problem hiding this comment.
Documentation for the terms should be added to the R level statistic documentation.
|
In general, this is not finished. I've been experimenting with this and for example the first Y.Y parameter corresponds with (-1/sigma^2) so must always be negative. I believe this is equivalent to the log partition function being finite issue. For gaussRegression to implement gaussian errorred regression will need to think about constrained optimization I think. |
Perhaps having a good approach to pseudo-likelihood based starting values would help solve this. Getting reasonable starting values would kick off the estimation procedure away from the undefined parameter space. |
Better starting values do help. I need to think about this more also. I have it working, but there is some "interesting" behaviour currently. |
2 participants
Adding some statistics to allow for network regression i.e. gaussian continuous nodal covariates with the predictors given by other nodal covariates or the sum of nodal covariates of neighbors. Added some test stats also.