[jira] [Closed] (MADLIB-1429) DL - utility to load and delete custom Python functions

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

     [ https://issues.apache.org/jira/browse/MADLIB-1429?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Frank McQuillan closed MADLIB-1429.
-----------------------------------
    Resolution: Fixed

https://github.com/apache/madlib/pull/498

> DL - utility to load and delete custom Python functions
> -------------------------------------------------------
>
>                 Key: MADLIB-1429
>                 URL: https://issues.apache.org/jira/browse/MADLIB-1429
>             Project: Apache MADlib
>          Issue Type: Improvement
>            Reporter: Frank McQuillan
>            Priority: Major
>             Fix For: v1.18.0
>
>
> Many deep learning models require loss functions and metrics that are not part of the standard deep learning libraries.
> In order to pass along this object to Keras,  we want to create a table which the user will populate with a pickled object mapping to the custom function.  We can model this utility on http://madlib.apache.org/docs/latest/group__grp__keras__model__arch.html which loads models into a model arch table.
> As part of this story, we want to create a utility function with the following api:
> {code}
> load_custom_function(
> object_table VARCHAR,  -- output table
> object BYTEA,  -- bytea object for custom function
> name VARCHAR,  -- function name (must be unique)
> description VARCHAR  -- description (optional)
> )
> {code}
> Creates the following output table:
> {code}
> |id |	name         | description   | function    |
> |---|-----------------|---------------|-------------|
> |1  | keras_custom_fn | test function | \x80026...  |
> {code}
> As part of this story, we want a delete utility function with the following api:
> {code}
> delete_custom_function(
> object_table VARCHAR,  -- table containing the Python object
> id -- id of object to delete
> )
> {code}



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