Basic Parameters

Note that this is for learning an RDN. The basic version boosts a single conditional relational probability distribution.

-l : enable training (learning).
-i : enable testing (inference).
-noBoost : disable Boosting (i.e., learns a single relational regression tree).
-train <Training directory> : Path to the training directory in predicate logic format.
-test <Testing directory> : Path to the testing directory in predicate logic format format.
-model <Model directory> : Path to the directory with the stored models [or where they will be stored].

Default location: “Training directory”/models
-target <target predicates> : Comma separated list of predicates to be learned/inferred.
-trees <Number of trees> : Number of Boosting trees.

Default: 20. Ignored if -noBoost is set.
-step <Step Length> : Default step length for functional gradient.

Default: 1. Ignored if lineSearch is set in advanced parameters.
-modelSuffix <suffix> : All the trees/models are saved with this suffix appended to the file names.
-aucJarPath <path to auc.jar> : If this is not set, AUC values are not computed.
-testNegPosRatio <Negative/Positive ratio> : Ratio of negatives to positive for testing.

Default: 2. Set to -1 to disable sampling.

Try to follow along with what each of these are doing:

From the Getting Started tutorial:

java -jar BoostSRL.jar -l -combine -train train/ -target father -trees 10
java -jar BoostSRL.jar -i -model train/models -test test/ -target father -trees 10 -aucJarPath <Path to auc.jar>

(If auc.jar does not exist in the source code folder, download from http://mark.goadrich.com/programs/AUC/)

From the Boston Housing Dataset (notice the different classpath):

java -cp BoostSRL.jar edu.wisc.cs.will.Boosting.Regression.RunBoostedRegressionTrees -reg -l -train train/ -target medv -trees 20

java -jar BoostSRL.jar -l -train train/ -target infield_fauthor,infield_ftitle,infield_fvenue -trees 5