You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@mxnet.apache.org by "Denis Davydenko (JIRA)" <ji...@apache.org> on 2019/01/02 20:34:00 UTC

[jira] [Updated] (MXNET-428) Switch data type for NDArray creation

     [ https://issues.apache.org/jira/browse/MXNET-428?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Denis Davydenko updated MXNET-428:
----------------------------------
    Resolution: Duplicate
        Status: Done  (was: To Do)

When Float64 is available - this functionality will be available as well.

> Switch data type for NDArray creation
> -------------------------------------
>
>                 Key: MXNET-428
>                 URL: https://issues.apache.org/jira/browse/MXNET-428
>             Project: Apache MXNet
>          Issue Type: Improvement
>          Components: Apache MXNet Scala API
>            Reporter: Koen Dejonghe
>            Priority: Major
>
> [Scala]
> It should be possible to swap the data type from float to double for creation of NDArrays, in order to perform gradient checks. 
> For more information on gradient checks, see http://cs231n.github.io/neural-networks-3/#gradcheck
> As an idea, in ND4J (https://nd4j.org/) it is possible to globally set the data type by executing:
> Nd4j.setDataType(DataBuffer.Type.DOUBLE)
> Subsequent nd-arrays will be of type Double.



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

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@mxnet.apache.org
For additional commands, e-mail: issues-help@mxnet.apache.org