Error scoring conditional random forests

[Nathan Hugenberger]

I can't seem to get the scoring/prediction part of conditional random forests (cforest/party) to work properly in rattle.  When I build a traditional random forest and score a validation sample, it works perfectly fine.  But when I try to build a conditional forest and the score the validation data, I get this error below.  Any suggestions?  I think I am doing things exactly the same way in Rattle (using the weather.csv datafile), so I am not sure why things are not working.

 

An error occured in the following command:

crs$pr <- predict(crs$rf, na.omit(crs$dataset[crs$validate, c(crs$input)])).

 

The error message was:

Error in OOB && is.null(newdata) : invalid 'x' type in 'x && y'

Contact ...   Rattle 2.6.17

Thanks for the help.

 

[Graham Williams]

The predict command for a cforest from the party package seems to now
need OOB=TRUE supplied to it (not sure when this happened nor whether
it is intentional but still checking). If you copy the predict command
from the Rattle Log tab, and add OOB=TRUE, the commands should work.
I've included a workaround for the next rattle version.

Thanks for reporting the problem Nathan.

Regards,
Graham