Using tweedie distribution in Rulefit: how to specify tweedie_power?

[Stuart Romm]

I am using rulefit on an insurance dataset for which a tweedie distribution would be appropriate. However, with such a distribution, one needs to specify the variance_power of the desired tweedie estimation, a number in the open interval (1,2). But when I try specifying variance power as follows (in the R h2o package), I get an "unused parameter" error:

 rfit <- h2o.rulefit(y = response,

                    x = predictors,

                    training_frame=train,

                    weights_column="exposure",

                    min_rule_length = 2,

                    max_rule_length = 3,

                    max_num_rules = 20,

                    rule_generation_ntrees = 25,

                    distribution="tweedie",

                    tweedie_power=1.65,

                    algorithm = "GBM",

                    seed=1234)

The error is thrown whether I use "tweedie_power" (as one would with GBM) or "tweedie_variance_power" (as one would with GLM.) I would appreciate any advice on how to do this, or alternatively a confirmation that tweedie isn't currently implemented in  rulefit and thus the distribution=tweedie option shouldn't be used right now.

[Juan Garza]

Hey Stuart,

I noticed that you haven't received a response for this yet. Let me invite you to post that question on our community forums so that others can see it and interact with it. You can join here: http://community.h2o.ai

See in the forums!

Juan Garza

Director of Community, H2O.ai