topic Re: Tune Gradient Boosting Node in SAS Data Science

Tune Gradient Boosting Node

aha123 — Wed, 27 Jul 2016 20:30:23 GMT

For my dataset, I run Regression, Neural Network, Decision Trees & Gradient Boosting Nodes. GB node always has the highest MSE, even higher than Regression. Even after I changed N interations to 200, it is still the worst one. Could you suggest what other parameters I can try out to improve the outcome? I run my dataset on R & Python, GB from both give very good results. Don't know what is wrong with SAS's GB node. My problem is of regression, not of classification.

Re: Tune Gradient Boosting Node

WendyCzika — Fri, 29 Jul 2016 16:56:31 GMT

One thing to try would be using a smaller value for Leaf Fraction - could be too high to find splits.

Re: Tune Gradient Boosting Node

Babloo — Mon, 01 Aug 2016 14:26:53 GMT

May I know what is leaf fraction and why we need to use a smaller for it?

Re: Tune Gradient Boosting Node

Babloo — Wed, 06 Feb 2019 07:05:38 GMT

May I know what is leaf fraction in Gradient Boosting?

Re: Tune Gradient Boosting Node

WendyCzika — Wed, 06 Feb 2019 13:56:50 GMT

Use the Leaf Fraction property to specify the smallest number of training observations a new branch may have, expressed as the proportion of the number N of available training observations in the data. N may be less than the total number of observations in the data set because observations with a missing target value are excluded.

So setting this to a smaller value grows larger trees.