You are viewing a plain text version of this content. The canonical link for it is here.

Posted to user@spark.apache.org by Pralabh Kumar <pr...@gmail.com> on 2017/06/16 02:30:30 UTC

Re: featureSubsetStrategy parameter for GradientBoostedTreesModel

Hi everyone

Currently GBT doesn't expose featureSubsetStrategy as exposed by Random
Forest.

.
GradientBoostedTrees in Spark have hardcoded feature subset strategy to
"all" while calling random forest in  DecisionTreeRegressor.scala

val trees = RandomForest.run(data, oldStrategy, numTrees = 1,
featureSubsetStrategy = "all",

It should provide functionality to the user to set featureSubsetStrategy
("auto", "all" ,"sqrt" , "log2" , "onethird") ,
the way random forest does.

This will help GBT to have randomness at feature level.

Jira SPARK-20199 <https://issues.apache.org/jira/browse/SPARK-20199>

Please let me know , if my understanding is correct.

Regards
Pralabh Kumar

On Fri, Jun 16, 2017 at 7:53 AM, Pralabh Kumar <pr...@gmail.com>
wrote:

> Hi everyone
>
> Currently GBT doesn't expose featureSubsetStrategy as exposed by Random
> Forest.
>
> .
> GradientBoostedTrees in Spark have hardcoded feature subset strategy to
> "all" while calling random forest in  DecisionTreeRegressor.scala
>
> val trees = RandomForest.run(data, oldStrategy, numTrees = 1,
> featureSubsetStrategy = "all",
>
>
> It should provide functionality to the user to set featureSubsetStrategy
> ("auto", "all" ,"sqrt" , "log2" , "onethird") ,
> the way random forest does.
>
> This will help GBT to have randomness at feature level.
>
> Jira SPARK-20199 <https://issues.apache.org/jira/browse/SPARK-20199>
>
> Please let me know , if my understanding is correct.
>
> Regards
> Pralabh Kumar
>