This plugin provides the ability to extract a set of interpretable rules from a Random Forest. The approach is based on the article “Learning accurate and interpretable models based on regularized random forests regression“.
Plugin Information
Version | 0.5.6 |
---|---|
Author | Dataiku (Pierre Pfennig & Du Phan) |
Released | 2018-08-01 |
Last updated | 2018-08-01 |
How To Use
You first need to set some general parameters: target, type of prediction, metric to optimize, performance drop threshold …
Next there are two parameters for the algorithm:
- Number of clauses in a rule: the longer the rule, the more accurate it is, but the less generalizable it is. Behind the scene, this is the max_depth of the random forest.
- Regularisation parameters: to filter out noisy rules, we use Lasso. This is the alpha parameter of Lasso.
Finally, using the min support parameter, you can also choose to precise that all rules below a certain support threshold are not interesting.