Uploaded image for project: 'Spark'
  1. Spark
  2. SPARK-10668

Use WeightedLeastSquares in LinearRegression with L2 regularization if the number of features is small

    XMLWordPrintableJSON

Details

    • New Feature
    • Status: Resolved
    • Critical
    • Resolution: Fixed
    • None
    • 1.6.0
    • ML
    • None

    Description

      If the number of features is small (<=4096) and the regularization is L2, we should use WeightedLeastSquares to solve the problem rather than L-BFGS. The former requires only one pass to the data.

      Attachments

        Issue Links

          blocks

          Sub-task - The sub-task of the issue SPARK-9836 Provide R-like summary statistics for ordinary least squares via normal equation solver

          • Critical - Crashes, loss of data, severe memory leak.
          • Resolved
          is related to

          Improvement - An improvement or enhancement to an existing feature or task. SPARK-11207 Add test cases for solver selection of LinearRegression as followup.

          • Major - Major loss of function.
          • Resolved
          relates to

          New Feature - A new feature of the product, which has yet to be developed. SPARK-9834 Normal equation solver for ordinary least squares

          • Major - Major loss of function.
          • Resolved
          links to

          Pull request #8884 [Github] Pull Request #8884 (Lewuathe)

          Activity

            People

              lewuathe Kai
              mengxr Xiangrui Meng
              DB Tsai DB Tsai
              Votes:
              0 Vote for this issue
              Watchers:
              6 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: