[jira] [Commented] (FLINK-1716) Add CoCoA algorithm to flink-ml
[ https://issues.apache.org/jira/browse/FLINK-1716?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14392405#comment-14392405 ] ASF GitHub Bot commented on FLINK-1716: --- Github user vasia commented on the pull request: https://github.com/apache/flink/pull/545#issuecomment-88834217 Hi @tillrohrmann! Thanks for the heads up :) The failing test is `DegreesWithExceptionITCase` and it seems that it fails because of not throwing the excepted exception type. We had a discussion about this in #475 and if I get it correctly, we can't be sure which exception will be reported to the client, because of the job cancelling. Thus, we removed the check of the exception type and message and only checked for failure. However, I see you added some exception checks in 36b762e. So, I'm not sure what we're supposed to do with this test now, so that it doesn't fail o.O Add CoCoA algorithm to flink-ml --- Key: FLINK-1716 URL: https://issues.apache.org/jira/browse/FLINK-1716 Project: Flink Issue Type: New Feature Components: Machine Learning Library Reporter: Till Rohrmann Assignee: Till Rohrmann Labels: ML Add the communication efficient distributed dual coordinate ascent algorithm to the flink machine learning library. See [CoCoA|http://arxiv.org/pdf/1409.1458.pdf] for the implementation details. I propose to first implement it with hinge loss and l2-norm. This way, it will allow us to train SVMs in parallel. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (FLINK-1716) Add CoCoA algorithm to flink-ml
[ https://issues.apache.org/jira/browse/FLINK-1716?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14392418#comment-14392418 ] ASF GitHub Bot commented on FLINK-1716: --- Github user tillrohrmann commented on the pull request: https://github.com/apache/flink/pull/545#issuecomment-88836418 Oh ok, I was not aware of #475. I just added the additional checks because of the fail message in the try block that you expect a ```NoSuchElementException```. If this is not a case, then we can just adapt the fail message and revert the additional checks I've introduced. Sorry for screwing up your test cases. Add CoCoA algorithm to flink-ml --- Key: FLINK-1716 URL: https://issues.apache.org/jira/browse/FLINK-1716 Project: Flink Issue Type: New Feature Components: Machine Learning Library Reporter: Till Rohrmann Assignee: Till Rohrmann Labels: ML Add the communication efficient distributed dual coordinate ascent algorithm to the flink machine learning library. See [CoCoA|http://arxiv.org/pdf/1409.1458.pdf] for the implementation details. I propose to first implement it with hinge loss and l2-norm. This way, it will allow us to train SVMs in parallel. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (FLINK-1716) Add CoCoA algorithm to flink-ml
[ https://issues.apache.org/jira/browse/FLINK-1716?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14392364#comment-14392364 ] ASF GitHub Bot commented on FLINK-1716: --- Github user tillrohrmann commented on the pull request: https://github.com/apache/flink/pull/545#issuecomment-88820531 Travis failed because of some Gelly test case. Travis running on my own repo passed, though. I'll merge the PR then. Add CoCoA algorithm to flink-ml --- Key: FLINK-1716 URL: https://issues.apache.org/jira/browse/FLINK-1716 Project: Flink Issue Type: New Feature Components: Machine Learning Library Reporter: Till Rohrmann Assignee: Till Rohrmann Labels: ML Add the communication efficient distributed dual coordinate ascent algorithm to the flink machine learning library. See [CoCoA|http://arxiv.org/pdf/1409.1458.pdf] for the implementation details. I propose to first implement it with hinge loss and l2-norm. This way, it will allow us to train SVMs in parallel. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (FLINK-1716) Add CoCoA algorithm to flink-ml
[ https://issues.apache.org/jira/browse/FLINK-1716?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14386820#comment-14386820 ] ASF GitHub Bot commented on FLINK-1716: --- GitHub user tillrohrmann opened a pull request: https://github.com/apache/flink/pull/545 [FLINK-1716] Adds CoCoA algorithm This PR adds the CoCoA algorithm using l2-norm and hinge-loss functions. Thus, it can be used to train soft-margin SVM. This PR is based on #539 and #543. You can merge this pull request into a Git repository by running: $ git pull https://github.com/tillrohrmann/flink cocoa Alternatively you can review and apply these changes as the patch at: https://github.com/apache/flink/pull/545.patch To close this pull request, make a commit to your master/trunk branch with (at least) the following in the commit message: This closes #545 commit 4c18940bf14f376cdb339d908324e5f2cd4593ad Author: Till Rohrmann trohrm...@apache.org Date: 2015-03-25T14:27:58Z [FLINK-1718] [ml] Adds sparse matrix and sparse vector types commit f3d021febf0e7796a1f250c2e693d7f9dcbc36e1 Author: Till Rohrmann trohrm...@apache.org Date: 2015-03-26T16:44:17Z [ml] Adds convenience functions for Breeze matrix/vector conversion [ml] Adds breeze to flink-dist LICENSE file [ml] Optimizes sanity checks in vector/matrix accessors [ml] Fixes scala check style error with missing whitespaces before and after + [ml] Fixes DenseMatrixTest commit be8ca43b5f11c789b2acfe38127ed542cdea3cd3 Author: Till Rohrmann trohrm...@apache.org Date: 2015-03-28T17:31:02Z [FLINK-1717] [ml] Adds support to directly read libSVM and SVMLight files commit 850d5880be0c7e484fa14e92b8af61dd5ceb6d4f Author: Till Rohrmann trohrm...@apache.org Date: 2015-03-12T15:52:45Z [FLINK-1716] [ml] Adds CoCoA algorithm [ml] Adds web documentation and code comments to CoCoA [ml] Adds comments Add CoCoA algorithm to flink-ml --- Key: FLINK-1716 URL: https://issues.apache.org/jira/browse/FLINK-1716 Project: Flink Issue Type: New Feature Components: Machine Learning Library Reporter: Till Rohrmann Assignee: Till Rohrmann Labels: ML Add the communication efficient distributed dual coordinate ascent algorithm to the flink machine learning library. See [CoCoA|http://arxiv.org/pdf/1409.1458.pdf] for the implementation details. I propose to first implement it with hinge loss and l2-norm. This way, it will allow us to train SVMs in parallel. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
