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

Posted to dev@singa.apache.org by "wangwei (JIRA)" <ji...@apache.org> on 2016/06/27 12:23:52 UTC

[jira] [Created] (SINGA-208) Add DataIter base class and a simple implementation

wangwei created SINGA-208:
-----------------------------

             Summary: Add DataIter base class and a simple implementation
                 Key: SINGA-208
                 URL: https://issues.apache.org/jira/browse/SINGA-208
             Project: Singa
          Issue Type: New Feature
            Reporter: wangwei


Data iterator is typically used for iterative algorithms like SGD to load training data (e.g., loading data from disk to main memory, and then to GPU).
The Data iterator could combine Reader (SINGA-202) instances and decoder (SINGA-200) instances to prefetch data.
It generates a mini-batch of training data at one time, e.g., using the Next() function.

This ticket would add the base DataIter class and implement a simple sub-class to read cifar10 dataset and generates the mini-batches.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)