ignite-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Artem Malykh (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (IGNITE-10955) [ML] Migrate boosting implementation to sequential trainers composition combinator
Date Tue, 29 Jan 2019 11:11:00 GMT

    [ https://issues.apache.org/jira/browse/IGNITE-10955?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16754883#comment-16754883
] 

Artem Malykh commented on IGNITE-10955:
---------------------------------------

Current implementation of boosting appears to be more readable in the state as it is currently
implemented. The problem with migration is that training in boosting is sequential while models
composition is parallel with weighted aggregator. It is possible to extend sequential trainer
composition to output some general composition with specification of this composition producer,
but for me it seems more cumbersome than it is currently implemented in GDB. But work that
has been done related to this feature and that maybe be helpful in future is extracted into
a separate ticket (IGNITE-111222).

> [ML] Migrate boosting implementation to sequential trainers composition combinator
> ----------------------------------------------------------------------------------
>
>                 Key: IGNITE-10955
>                 URL: https://issues.apache.org/jira/browse/IGNITE-10955
>             Project: Ignite
>          Issue Type: Improvement
>          Components: ml
>    Affects Versions: 2.8
>            Reporter: Artem Malykh
>            Assignee: Artem Malykh
>            Priority: Major
>
> There are two trainers composition primitives which are used in other ensemble training
methods (Bagging and Stacking) implementation. To unify implementation I suggest to rewrite
impl of boosting using these composition primitives as well.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Mime
View raw message