[jira] [Created] (MADLIB-988) Model parameter weighting

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

Frank McQuillan created MADLIB-988:
--------------------------------------

             Summary: Model parameter weighting
                 Key: MADLIB-988
                 URL: https://issues.apache.org/jira/browse/MADLIB-988
             Project: Apache MADlib
          Issue Type: New Feature
            Reporter: Frank McQuillan
         Attachments: Model_parameter_weighting.pdf

Summary

There are several instances where assigning weights to training samples or observations is desirable in order to portray known information about the data or handle situations where data quality varies.  For example, the training sample may have a disproportionate number of observations in certain classes, or the data may have been collected in a stratified manner with one strata having greater or lesser sampling intensity. In such cases, observations can be weighted to reflect the importance of each point in the fitted model.

References

[1] See requirements document authored by Pivotal data science team
(attached)




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