leaf_threshold the threshold for leaf. can be right max_right max or all . pruning_type right this is exactly what I describe above. We take all the right branches only the sign is use in the conditions in the leaf of which the share of the positive class is higher than leaf_threshold and class objects are greater than min_positive. pruning_type max right this is the same as right only an additional condition is impose on the branch we take the right branch if its leaf has the most objects of a positive class among.
The max and all
All leaves that satisfy the condition of the share and the minimum presence of objects. Simply put the right branch should not only be prolific for our class but also the most in the tree. to be even more effective than right. It additionally cuts off weak branches even Senegal WhatsApp Number List if they satisfy the criterion. Branch for max_right Branch for max_right pruning_type is higher than leaf_threshold we take the one in whose leaf there are the maximum objects of the class and more than min_positive. Those. its like max_right but without right. pruning_type all these are all branches of the tree in the leaf of which the share of the positive class is higher than leaf_threshold and class objects are greater than minpositive.
Features option only
If leaf_threshold and min_positive then these are simply all the branches of all trees. But all the branches of all the trees do not make Random Branch similar to Random Forest since Random . This type of pruning works by increasing the min_positive value. Using the min Argentina Phone Number List positive condition is not the same as using a tree growth limit min_samples_leaf. Using min_positive we work with an already constructe tree. And we dont take a branch if its leaf contains fewer objects of the positive class than the specifie min_positive. In general base on the experience of several tasks the most effective and stable option is max_right then right.