linear fraction tree ensemble

Hi, I’m just wondering if someone could explain how the linear fraction in the tree ensemble works. I am using the regression version of a learner and am finding success with the linear fraction, so I’m just curious how the fraction works. Also, would you happen to know how the absolute value option works? It seems a bit weird taking the absolute value of features for some type of sampling method.

Hi @Haseeb23

The linear fraction option in the attribute sampling is the percentage of columns that are available for a split / are potential splitting columns. Consider 1=100% and 0.5=50% of all columns, etc. In the configuration below (“Attribute Selection”) you can then determine if that applies for the whole tree or just for a tree node.

The absolute value option is similar, just that there you can determine the absolute number of columns that can be taken for a split. So if you type 5, it will consider 5 columns.

I hope that made it clearer to you :slight_smile:


This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.