Motivation

For H2O models trained with an offset column it is necessary to provide offset column for scoring as well otherwise H2O throws an error. Even though in most cases omission of the offset column is likely a mistake there are use cases when user doesn't have the information at scoring time and still wants to make a prediction and post-process the prediction later.

Solution

If offset column is not available at scoring time, user needs to explicitly create a zero offset column. This will make the algorithms behave as if the offset was not specified.

Example:

library(h2o)
h2o.init(nthreads = -1)

# TODO: add an example

References

• Documentation of offset column: http://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-science/algo-params/offset_column.html

JIRA Issue Migration Info

Jira Issue: TN-12
Assignee: Michal Kurka
Reporter: Michal Kurka
State: In Progress