[jira] [Created] (MAHOUT-1990) Implement Multilayer Perceptron

ASF GitHub Bot (JIRA)

2017-05-28 16:15:05 UTC

[ https://issues.apache.org/jira/browse/MAHOUT-1990?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16027850#comment-16027850 ]

ASF GitHub Bot commented on MAHOUT-1990:
----------------------------------------

GitHub user rawkintrevo opened a pull request:

https://github.com/apache/mahout/pull/323

MAHOUT-1990 [WIP] Add Multilayer Perceptron

### Purpose of PR:
Implements multi-layer perceptron style neural network

Also introduces some SGD functionality (namely learning strategies) (will, doesn't yet)

Also introduces classifier trait

### Important ToDos
Please mark each with an "x"
- [x] A JIRA ticket exists (if not, please create this first)[https://issues.apache.org/jira/browse/ZEPPELIN/]
- [x] Title of PR is "MAHOUT-XXXX Brief Description of Changes" where XXXX is the JIRA number.
- [ ] Created unit tests where appropriate
- [ ] Added licenses correct on newly added files
- [x] Assigned JIRA to self
- [ ] Added documentation in scala docs/java docs, and to website
- [ ] Successfully built and ran all unit tests, verified that all tests pass locally.

Does this change break earlier versions?
**yes** refactored linear regression to new sub-package, and created nonlinear package.

Is this the beginning of a larger project for which a feature branch should be made?
... possibly, but I think I will have it all working in short enough order that it can stay on mine- however I am open to this idea if someone else wants to chip in.

You can merge this pull request into a Git repository by running:

$ git pull https://github.com/rawkintrevo/mahout mahout-1990

Alternatively you can review and apply these changes as the patch at:

https://github.com/apache/mahout/pull/323.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 #323

----
commit d15a9c2730888aeb0973c173ccdaeb12957b2c53
Author: rawkintrevo <***@gmail.com>
Date: 2017-05-25T06:49:10Z

initial incoreMLP work

commit ab2b1213b76f1f94297a617d7f465479825d7377
Author: rawkintrevo <***@gmail.com>
Date: 2017-05-27T15:48:07Z

Distributed MLP

commit 226a884c8e7d520cd4656ab9d8cfd71647a84b0f
Author: rawkintrevo <***@gmail.com>
Date: 2017-05-28T04:30:45Z

Distributed works now

commit 7ea981e53af692a43bc4b75c176b46b7ff6ce328
Author: rawkintrevo <***@gmail.com>
Date: 2017-05-28T14:05:47Z

Distributed works now

commit c32585be8d0260c0a6000ebd6944c0b1a2c95595
Author: rawkintrevo <***@gmail.com>
Date: 2017-05-28T16:03:24Z

Added regression and classifier wrappers

----

Post by Trevor Grant (JIRA)
Implement Multilayer Perceptron
-------------------------------
Key: MAHOUT-1990
URL: https://issues.apache.org/jira/browse/MAHOUT-1990
Project: Mahout
Issue Type: Improvement
Components: Algorithms
Affects Versions: 0.13.2
Reporter: Trevor Grant
Assignee: Trevor Grant
Following strategy
It should-
1. implement incoreMLPs which can be 'plugged together' for purposes of back propegation (this makes for easy extension into more complex networks)
2. implement a common distributed MLP which maps out incoreMLPs and then averages parameters
3. regression and classifier wrappers around the base MLP to reduce duplication of code
4. would be nice to make distributed and incore neural network 'trait' for consistent API across all future neural networks.

--
This message was sent by Atlassian JIRA
(v6.3.15#6346)