[jira] [Commented] (SOLR-12780) Add support for Leaky ReLU and TanH activations in LTR contrib module

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

[jira] [Commented] (SOLR-12780) Add support for Leaky ReLU and TanH activations in LTR contrib module

JIRA jira@apache.org

    [ https://issues.apache.org/jira/browse/SOLR-12780?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16646782#comment-16646782 ]

Christine Poerschke commented on SOLR-12780:
--------------------------------------------

Thanks [~kamulau] for opening this ticket with a patch to add two additional activation functions!

Just attached slightly revised patch, the main difference is the (proposed) use of {{Math.tanh}} instead of the
{code}
(Math.exp(in) - Math.exp(-in))/(Math.exp(in) + Math.exp(-in))
{code}
formula - what do you think? Initially I'd wondered about the benefits or otherwise of reducing the number of {{Math.exp}} calls and then your SOLR-12785 patch made me wonder if [Apache Commons Math|http://commons.apache.org/proper/commons-math/javadocs/api-3.6.1/index.html] has activation functions and that then led to the discovery that {{Math.tanh}} exists in Java itself!

> Add support for Leaky ReLU and TanH activations in LTR contrib module
> ---------------------------------------------------------------------
>
>                 Key: SOLR-12780
>                 URL: https://issues.apache.org/jira/browse/SOLR-12780
>             Project: Solr
>          Issue Type: New Feature
>      Security Level: Public(Default Security Level. Issues are Public)
>          Components: contrib - LTR
>            Reporter: Kamuela Lau
>            Priority: Minor
>              Labels: ltr
>         Attachments: SOLR-12780.patch, SOLR-12780.patch
>
>
> Add support for Leaky ReLU and TanH activation functions in NeuralNetworkModel.



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

---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]