[jira] [Commented] (MADLIB-1420) Support to model one dimension data in keras

["Frank McQuillan (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MADLIB-1420?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17069023#comment-17069023 ] 

Frank McQuillan commented on MADLIB-1420:
-----------------------------------------

Actually you *can* use MADlib with Keras/TF for data that is not images.

For example see 
https://github.com/apache/madlib-site/blob/asf-site/community-artifacts/Deep-learning/MADlib-Keras-model-selection-MLP-v1.ipynb
(multi-model training supported in 1.17+)
&
https://github.com/apache/madlib-site/blob/asf-site/community-artifacts/Deep-learning/MADlib-Keras-MLP-v2.ipynb
(single model training in supported 1.16+)

which use the Iris data set as a non-image example.  Just ensure that you create a flat array for your feature vector and mini-batch it in the usual way, and make sure your neural net input shape is set up properly to accept the feature array.

The cols to vec utility function may be of help in creating a feature vector from columns of a table
http://madlib.apache.org/docs/latest/group__grp__cols2vec.html
https://github.com/apache/madlib-site/blob/asf-site/community-artifacts/Utilities/Column-vector-operations-v1.ipynb

Please let us know if you need any additional information on this.



> Support to model one dimension data in keras
> --------------------------------------------
>
>                 Key: MADLIB-1420
>                 URL: https://issues.apache.org/jira/browse/MADLIB-1420
>             Project: Apache MADlib
>          Issue Type: New Feature
>            Reporter: Hubert Zhang
>            Priority: Major
>
> Currently, MADlib with keras focus on image processing.
> Is there any way to handle one dimension data?



--
This message was sent by Atlassian Jira
(v8.3.4#803005)