GitHub user srowen opened a pull request:
https://github.com/apache/spark/pull/14949
[SPARK-17057] [ML] ProbabilisticClassifierModels' prediction more
reasonable with multi zero thresholds
## What changes were proposed in this pull request?
See related discussion at https://github.com/apache/spark/pull/14643
This actually changes more than what the original JIRA encompassed, but
does propose a more reasonable (?) and deterministic result in this and other
corner cases.
Revise semantics of ProbabilisticClassifierModel thresholds so that classes
can only be predicted if they exceed their threshold (meaning no class may be
predicted), and otherwise ordering by highest probability, then lowest
threshold, then by class index.
## How was this patch tested?
Existing and new unit tests.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/srowen/spark SPARK-17057.2
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/spark/pull/14949.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #14949
commit 08dbe43bbf961186ee94432a13f9f6cfc221e4a6
Author: Sean Owen
Date: 2016-09-03T14:26:26Z
Revise semantics of ProbabilisticClassifierModel thresholds so that classes
can only be predicted if they exceed their threshold (meaning no class may be
predicted), and otherwise ordering by highest probability, then lowest
threshold, then by class index
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
-
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org