h2o.h function

Calculates Friedman and Popescu's H statistics, in order to test for the presence of an interaction between specified variables in h2o gbm and xgb models. H varies from 0 to 1. It will have a value of 0 if the model exhibits no interaction between specified variables and a correspondingly larger value for a stronger interaction effect between them. NaN is returned if a computation is spoiled by weak main effects and rounding errors.

This statistic can be calculated only for numerical variables. Missing values are supported.

h2o.h(model, frame, variables)

Arguments

• model: A trained gradient-boosting model.
• frame: A frame that current model has been fitted to.
• variables: Variables of the interest.

Details

See Jerome H. Friedman and Bogdan E. Popescu, 2008, "Predictive learning via rule ensembles", Ann. Appl. Stat. 2:916-954, http://projecteuclid.org/download/pdfview_1/euclid.aoas/1223908046, s. 8.1.

Examples

## Not run:

library(h2o)
h2o.init()
prostate.hex <- h2o.importFile(
       "https://s3.amazonaws.com/h2o-public-test-data/smalldata/logreg/prostate.csv",
        destination_frame="prostate.hex"
        )
prostate.hex$CAPSULE <- as.factor(prostate.hex$CAPSULE)
prostate.hex$RACE <- as.factor(prostate.hex$RACE)
prostate.h2o <- h2o.gbm(x = 3:9, y = "CAPSULE", training_frame = prostate.hex, 
distribution = "bernoulli", ntrees = 100, max_depth = 5, min_rows = 10, learn_rate = 0.1)
h_val <- h2o.h(prostate.h2o, prostate.hex, c('DPROS','DCAPS'))
## End(Not run)