Open fabian-s opened 8 years ago
Hi Fabian,
awesome!
Some notes/thoughts:
Risk0
will be 0 and a negative importance is set for the first base-learner.frequence
Here is how it looks for my data:
RelRisk(zero)
RelRisk(zero, ohne1= FALSE)
and to compare my ggplot
solution:
zero_importance.pdf
Almond Stöcker und Tobi Kühn have started to write a function to extract the amount of risk reduction contributed by each base-learner from a fitted
mboost
model. We'd like to see this included in mboost, with visualisation options, if possible, as it seems to answer a question that comes up frequently in postprocessing and interpreting boosting fits: which base-learners are the most important for the fit? See code below.Any input on what such a function should or should not do would be highly appreciated!