[ https://issues.apache.org/jira/browse/FLINK-1993?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14570972#comment-14570972 ]
ASF GitHub Bot commented on FLINK-1993: --------------------------------------- Github user tillrohrmann commented on a diff in the pull request: https://github.com/apache/flink/pull/760#discussion_r31632765 --- Diff: flink-staging/flink-ml/src/main/scala/org/apache/flink/ml/pipeline/Predictor.scala --- @@ -35,7 +35,7 @@ import org.apache.flink.ml.common.{FlinkMLTools, ParameterMap, WithParameters} * * @tparam Self Type of the implementing class */ -trait Predictor[Self] extends Estimator[Self] with WithParameters with Serializable { +trait Predictor[Self] extends Estimator[Self] with WithParameters { --- End diff -- The `Estimator` types should never be shipped to the TaskManager. Usually they contain some state in the form of other `DataSets` which aren't serializable anyways. However, the operations have to be `Serializable`, because they can contain values which are referenced by the Flink operations. > Replace MultipleLinearRegression's custom SGD with optimization framework's > SGD > ------------------------------------------------------------------------------- > > Key: FLINK-1993 > URL: https://issues.apache.org/jira/browse/FLINK-1993 > Project: Flink > Issue Type: Task > Components: Machine Learning Library > Reporter: Till Rohrmann > Assignee: Till Rohrmann > Priority: Minor > Labels: ML > Fix For: 0.9 > > > The current implementation of MultipleLinearRegression uses a custom SGD > implementation. Flink's optimization framework also contains a SGD optimizer > which should replace the custom implementation once the framework is merged. -- This message was sent by Atlassian JIRA (v6.3.4#6332)