[
https://issues.apache.org/jira/browse/MXNET-1114?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Lin Yuan updated MXNET-1114:
----------------------------
Description:
-error: currently each GPU is running validation on only a portion of the
validation dataset, so validation accuracy trails training accuracy a lot at 90
epochs (8% vs. 75%)
-fix: change it so each GPU validates entire validation dataset
> allreduce validation results
> ----------------------------
>
> Key: MXNET-1114
> URL: https://issues.apache.org/jira/browse/MXNET-1114
> Project: Apache MXNet
> Issue Type: Bug
> Components: Apache MXNet Backend
> Reporter: Lin Yuan
> Priority: Major
>
> -error: currently each GPU is running validation on only a portion of the
> validation dataset, so validation accuracy trails training accuracy a lot at
> 90 epochs (8% vs. 75%)
> -fix: change it so each GPU validates entire validation dataset
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
