I'm running a bunch of "Gradient Boosting Method" nodes each with different parameters.
I would like to export the variable importance table of each model and append all in a unique dataset in order to get average importance values.
When you run only a single Gradient Boosting node, the Results browser includes a Variable Importance table. This table is not displayed when you run the Gradient Boosting node in group processing. However, a data set called boost_importance_loop is created, and you can access this data set to see the variable importance.
If you are running Gradient Boosting inside a Group Processing loop, you will need to write out the boost_importance_loop after each loop, but there is then no standard way to combine those importance measures into a single importance measure. If you have 10 loops with 10 different variable importance values, what is the true importance? Therefore, no overall variable importance table is produced in this situation.
Hope this helps!
Doug
... View more