Frank McQuillan created MADLIB-1451:
---------------------------------------
Summary: DL: Improve output of predict
Key: MADLIB-1451
URL: https://issues.apache.org/jira/browse/MADLIB-1451
Project: Apache MADlib
Issue Type: Improvement
Components: Deep Learning
Reporter: Frank McQuillan
Fix For: v1.18.0
**Story**
Applies to
https://madlib.apache.org/docs/latest/group__grp__keras__run__model__selection.html#keras_predict
&
https://madlib.apache.org/docs/latest/group__grp__keras.html#keras_predict
&
https://madlib.apache.org/docs/latest/group__grp__keras.html#keras_predict_byom
Idea:
Only want to run predict once and operate on the output table. You should not
have to run predict over again to get a different format (e.g, top 1 vs. top
5), which would be inefficient. To do this we change the meaning of the param
`pred_type` :
{code}
pred_type (optional)
TEXT, default: 'all'. Type of output desired, where 'all' gives predictions for
all classes and their associated
probabilities. Alternatively, you can specify a filter for top n or minimum
probability. For top n, use
an INTEGER > 0 to indicate the top ranked probabilities to output. For minimum
probability, specify
a REAL value between 0.0 and 1.0 for the cutoff.
{code}
table output format
{code}
id | class | probability | rank
----+-------------------+---------------+----------
2 | Iris-setosa | 0.8704131 | 1
2 | Iris-versicolor | 0.09302262 | 2
2 | Iris-virginica | 0.036564212 | 3
9 | Iris-virginica | 0.7704131 | 1
9 | Iris-versicolor | 0.19302262 | 2
9 | Iris-setosa | 0.136564212 | 3
etc.
{code}
**Open questions**
1) Will the above work OK with predict BYOM?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
