[ 
https://issues.apache.org/jira/browse/GRIFFIN-205?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16648899#comment-16648899
 ] 

Lionel Liu commented on GRIFFIN-205:
------------------------------------

In my opinion, the "total", "miss" and "matched" counts in accuracy measure 
results are raw metrics, they could be aggregated in later calculation. But the 
"matchedFraction" field is a calculated metrics, which could not be used in 
later aggregation. Combining these metrics in the same metric value might 
mislead the users.
I think it's OK to add "matchedFraction" field in accuracy metrics, but we need 
to clarify the difference between the count metric and fraction metric in 
document.
BTW, if we add this field in batch mode, it would be better to keep the 
consistency in streaming mode as well.

> Accuracy measure check should provide matchedFraction to store
> --------------------------------------------------------------
>
>                 Key: GRIFFIN-205
>                 URL: https://issues.apache.org/jira/browse/GRIFFIN-205
>             Project: Griffin (Incubating)
>          Issue Type: Improvement
>          Components: accuracy-batch, accuracy-real-time
>    Affects Versions: 0.3.1-incubating
>            Reporter: Artem Shutak
>            Assignee: Artem Shutak
>            Priority: Major
>
> Currently, {{accuracy}} measure results contains "total", "miss" and 
> "matched" counts.
> As a result, It's hard to analyze accuracy fraction based on results stored 
> in ElasticSearch, because ElasticSearch does not provide straight forward 
> capability to get "field divided by field" query results.
> {{Accuracy}} measure results should also contain {{matchedFraction}} field. 



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to