GitHub user brkyvz opened a pull request:
https://github.com/apache/spark/pull/2451
[WIP][SPARK-1486][MLlib] Multi Model Training with Gradient Descent
**Note: This is still a work in progress**
This is the first of the pull requests to support multi-model training in
MLlib. It batches examples and trains multiple models with different
regularization parameters and step sizes all at once using Matrix-Matrix
multiplication. It uses Native BLAS when the data matrix is dense, and uses
sparse matrices as much as possible for both better memory utilization and
performance (I will post performance results in the comments).
This is a HUGE Pull Request, therefore I'm posting this now. It is not
finished, docs need to be updated, code can be somewhat cleaned up for ease of
understanding. I'm posting this now so that users can comment and make
suggestions along the way.
Most of the PR consists of adding additional Local Matrix operations for
the calculation of gradients and losses.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/brkyvz/spark SPARK-1486
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/spark/pull/2451.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 #2451
----
commit 5138d3f220efb8ec4cc7fc944112497ab0c8c50c
Author: Burak <[email protected]>
Date: 2014-09-05T20:34:24Z
[SPARK-3418][MLlib] Sparse Matrix support and additional native BLAS
operations added
commit 4362ff1d6b79604b9919da1d550ed466021b8f5c
Author: Burak <[email protected]>
Date: 2014-09-05T20:51:45Z
[SPARK-3418][MLlib] Matrix unit tests expanded with indexing and updating
commit 8dcb7632093fc2d87a626b85e08ddabcbf119543
Author: Burak <[email protected]>
Date: 2014-09-05T21:58:55Z
[SPARK-3418] Fixed Scala-style errors
commit 41b2da30797e3da011e3261f8c2b89b9c1608d23
Author: Burak <[email protected]>
Date: 2014-09-06T01:30:29Z
[SPARK-3418] Fixed failing Matrix unit test
commit 56d7c85a1d58ef68c31208e062a4458e170111d3
Author: Burak <[email protected]>
Date: 2014-09-06T02:55:11Z
[SPARK-3418] Fixed style issues and added documentation for methods
commit 848406c1b6779eea9bbcf1dd582b541db46d7dad
Author: Burak <[email protected]>
Date: 2014-09-06T04:43:00Z
[SPARK-3418] Fixed one more style issue
commit eeb13ebda3223eb2f5fc36e08303e5b33d76de96
Author: Burak <[email protected]>
Date: 2014-09-09T00:55:40Z
[SPARK-3418] Code review comments addressed and multiplication further
optimized
commit a85ccb712d83d20b178b40c32bd473d7d018a88f
Author: Burak <[email protected]>
Date: 2014-09-09T07:22:09Z
[SPARK-3418] New code review comments addressed
commit d510c8f940faee3bdb2b00306f150bc99630396b
Author: Burak <[email protected]>
Date: 2014-09-14T04:07:59Z
[SPARK-3418] Squashed missing alpha bug.
commit 418def8e940b93f0d24e9b3158ecc0e130d16a83
Author: Burak <[email protected]>
Date: 2014-09-17T23:53:03Z
sealed traits Vector and Matrix
commit f79db9c0d82ceea41d59972e3bde9fa2a17b6112
Author: Burak <[email protected]>
Date: 2014-09-18T06:45:08Z
9/17 comments addressed
commit d16268496cefe06d8818b20721eede56a9de41a2
Author: Burak <[email protected]>
Date: 2014-09-18T18:55:28Z
[SPARK-3418] Fixed MiMa compatibility issues (excluded from check)
commit 272feb9f63517c52e1991988b4fbd8869a992dc4
Author: Burak <[email protected]>
Date: 2014-09-18T20:29:26Z
really fixed MiMa issue
commit 5e7d74408fd5f4e521f4e3a7e94a289d59454913
Author: Burak <[email protected]>
Date: 2014-09-18T21:16:05Z
[WIP][SPARK-1486][MLlib] Initial commit for multi-model training
----
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
