[ https://issues.apache.org/jira/browse/SPARK-15947?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15334725#comment-15334725 ]
Xiangrui Meng commented on SPARK-15947: --------------------------------------- Had an offline discussion with [~josephkb]. There would be lot of work to implement this feature and tests. A simpler choice is to ask users to manually convert the DataFrames at the beginning of the pipeline with tools implemented in SPARK-15945. Then we can update migration guide to include the error message and put this workaround there. So users can search on Google and find the solution. I'm closing this ticket. > Make pipeline components backward compatible with old vector columns > -------------------------------------------------------------------- > > Key: SPARK-15947 > URL: https://issues.apache.org/jira/browse/SPARK-15947 > Project: Spark > Issue Type: Sub-task > Components: ML, MLlib > Affects Versions: 2.0.0 > Reporter: Xiangrui Meng > Assignee: Xiangrui Meng > > After SPARK-15945, we should make ALL pipeline components accept old vector > columns as input and do the conversion automatically (probably with a warning > message), in order to smooth the migration to 2.0. > --Note that this includes loading old saved models.-- SPARK-16000 handles > backward compatibility in model loading. -- This message was sent by Atlassian JIRA (v6.3.4#6332) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org