Uh oh!
There was an error while loading. Please reload this page.
[SPARK-11171][SPARK-11237][SPARK-11241][ML] Try adding PMMLExportable to ML with KMeans - #9207
Conversation
SparkQA
commented
Oct 22, 2015
Test build #44104 has finished for PR 9207 at commit
|
368879f to
1749aecCompareSparkQA
commented
Oct 22, 2015
Test build #44105 has finished for PR 9207 at commit
|
SparkQA
commented
Oct 22, 2015
Test build #44111 has finished for PR 9207 at commit
|
… only (since actual PMML model evaluation is to be done through a spark-packages project and previous JIRA decided re-loading PMML is out of project scope)
holdenk
commented
Oct 23, 2015
cc @jkbradley |
SparkQA
commented
Oct 23, 2015
Test build #44241 has finished for PR 9207 at commit
|
holdenk
commented
Oct 29, 2015
re-ping @jkbradley |
1 similar comment
holdenk
commented
Nov 8, 2015
re-ping @jkbradley |
holdenk
commented
Nov 8, 2015
Also maybe @dbtsai if you have a chance to look at this that would be great. |
holdenk
commented
Dec 1, 2015
So I've updated this against master @jkbradley |
SparkQA
commented
Dec 1, 2015
Test build #46990 has finished for PR 9207 at commit
|
There was a problem hiding this comment.
Seems this is a copy-paste of org.apache.spark.mllib.pmml, should we deprecate the mllib one, and use the new one in ml package?
There was a problem hiding this comment.
I think that might be good, the main difference is this avoids using the factory implementation that the MLLib API was.
There was a problem hiding this comment.
(Uses the same public facing API as per the JIRA discussion re: lack of complaints from users with old API)
holdenk
commented
Dec 16, 2015
ping @jkbradley |
SparkQA
commented
Dec 30, 2015
Test build #48499 has finished for PR 9207 at commit
|
SparkQA
commented
Jan 11, 2016
Test build #49179 has finished for PR 9207 at commit
|
holdenk
commented
Jan 12, 2016
jenkins retest this please |
SparkQA
commented
Jan 12, 2016
Test build #49214 has finished for PR 9207 at commit
|
SparkQA
commented
Jan 21, 2016
Test build #49819 has finished for PR 9207 at commit
|
holdenk
commented
Jan 21, 2016
Jenkins retest this please |
SparkQA
commented
Jan 21, 2016
Test build #49837 has finished for PR 9207 at commit
|
SparkQA
commented
Jul 3, 2016
Test build #61683 has finished for PR 9207 at commit
|
SparkQA
commented
Jul 3, 2016
Test build #61684 has finished for PR 9207 at commit
|
SparkQA
commented
Jul 5, 2016
Test build #61784 has finished for PR 9207 at commit
|
holdenk
commented
Jul 5, 2016
HiveSubmit failures seem unrelated, jenkins retest this please. |
SparkQA
commented
Jul 6, 2016
Test build #61799 has finished for PR 9207 at commit
|
SparkQA
commented
Jul 18, 2016
Test build #62478 has finished for PR 9207 at commit
|
holdenk
commented
Jul 25, 2016
Now that 2.0 RC5 has passed - maybe it would be an OK time to revisit this? I think it can also be an ok base for adding more formats if we need to. |
SparkQA
commented
Aug 1, 2016
Test build #63086 has finished for PR 9207 at commit
|
SparkQA
commented
Aug 4, 2016
Test build #63193 has finished for PR 9207 at commit
|
SparkQA
commented
Aug 6, 2016
Test build #63295 has finished for PR 9207 at commit
|
holdenk
commented
Aug 8, 2016
ping @MLnick now that 2.0 is out for awhile. |
SparkQA
commented
Sep 9, 2016
Test build #65123 has finished for PR 9207 at commit
|
holdenk
commented
Nov 16, 2016
Ping @jkbradley / @MLnick - we are coming up on a year since we told people we would add PMML export to Spark in about a year so I figured I'd try and update this again. I'll update this PR tonight against the latest master. |
SparkQA
commented
Nov 16, 2016
Test build #68724 has finished for PR 9207 at commit
|
holdenk
commented
Sep 21, 2017
@MLnick: Do you have the bandwith to revisit this? I'm open to refactoring to a more plug-gable approach if we've got the review bandwidth for it. |
SparkQA
commented
Sep 23, 2017
Test build #82102 has finished for PR 9207 at commit
|
holdenk
commented
Mar 5, 2018
Closing this in favor of the approach taken in #19876 |
Add PMMLExportable to ML & use it for KMeans in ML pipeline. This is a bit different than the PMMLExportable in MLLib since the default trait doesn't go through a factory but instead depends on the model implementing it itself (and at the start that will mostly just be calling the parent model's method).
Still a WIP but wanted to see if anyone had comments about the slight change.