Uploaded image for project: 'Spark'
  1. Spark
  2. SPARK-3724

RandomForest: More options for feature subset size

    XMLWordPrintableJSON

    Details

    • Type: Improvement
    • Status: Resolved
    • Priority: Minor
    • Resolution: Fixed
    • Affects Version/s: None
    • Fix Version/s: 2.0.0
    • Component/s: MLlib
    • Labels:
      None

      Description

      RandomForest currently supports using a few values for the number of features to sample per node: all, sqrt, log2, etc. It should support any given value (to allow model search).

      Proposal: If the parameter for specifying the number of features per node is not recognized (as “all”, “sqrt”, etc.), then it will be parsed as a numerical value. The value should be either (a) a real value in [0,1] specifying the fraction of features in each subset or (b) an integer value specifying the number of features in each subset.

        Attachments

          Issue Links

            Activity

              People

              • Assignee:
                yongtang Yong Tang
                Reporter:
                josephkb Joseph K. Bradley
                Shepherd:
                Nicholas Pentreath
              • Votes:
                0 Vote for this issue
                Watchers:
                5 Start watching this issue

                Dates

                • Created:
                  Updated:
                  Resolved: