GitHub user fobeligi opened a pull request:
https://github.com/apache/incubator-samoa/pull/70
Samoa 72 New Boosting algorithm for VHT
This pull request provides BoostVHT algorithm, an online boosting algorithm
that leverages VHT algorithm as a base learner. BoostVHT design allows the
separation of the model boosting from its training, thus obtaining order of
magnitude speedups compared to the state-of-the-art.
Since a number of classes that are used by VHT algorithm are changed, we
performed tests to assure that VHT's performance was not affected, before this
pull request.
Below I refer briefly to a number of changes:
* `DecisionStump` class was initially imported from MOA in order to perform
some sanity checks. We kept the code in this PR.
* In `AttributeSplitSuggestion` class we change the public constructor to
set `merit = Double.NEGATIVE_INFINITY;`, instead of doing nothing, which didn't
affect VHT's performance in tests.
* Methods in `static class Builder` of `ModelAggregatorProcessor` are
changed to return a generic, as we extend the
`ModelAggregatorProcessor.Builder` class in `BoostMAProcessor` class.
* `LocalStatisticsProcessor` class was changed to also handle an
`AttributeSliceEvent`, an event that was introduced for BoostVHT algorithm.
* Classes that are used both from VHT and BoostVHT algorithms have a number
of access level changes.
Sure there are more changes than the ones I referred to above.
Îny feedback would be appreciated.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/fobeligi/incubator-samoa SAMOA-72
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/incubator-samoa/pull/70.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 #70
----
commit f3fbdefa6f942751b9d1de3a9a295d9ffc6b91c8
Author: fobeligi <[email protected]>
Date: 2017-01-17T18:19:06Z
[SAMOA-72] Initial topology and logic for BoostVHT alforithm.
commit 8ec73bf5a06c8b6f97f3c8e301ffa9fa0074f2b8
Author: Theodore Vasiloudis <[email protected]>
Date: 2017-04-20T14:11:34Z
[SAMOA-72] Optimize BoostVHT code with buffering and by introducing the
AttributeSliceEvent.
We bunch up all the attributes one local stats processor is responsible for
in an AttributeSliceEvent and send one message instead of creating one message
per single attribute.
commit 7469ee14360d6b069cebe5d9d698667fcbc5a0e3
Author: Theodore Vasiloudis <[email protected]>
Date: 2017-05-08T14:22:52Z
[SAMOA-72] Rebase on master branch and refactor BoostVHT code for pull
request.
----
---
