[SPARK-6332] [MLlib] compute calibration curve for binary classifier #5025
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@robert-dodier Thanks for the PR! I added a couple of clarification questions to the JIRA |
|
ok to test |
|
Can one of the admins verify this patch? |
| @@ -141,6 +141,7 @@ private[spark] abstract class ProbabilisticClassificationModel[ | |||
| * | |||
| * WARNING: Not all models output well-calibrated probability estimates! These probabilities | |||
| * should be treated as confidences, not precise probabilities. | |||
| * See also BinaryClassificationMetrics.calibration to assess calibration. | |||
There was a problem hiding this comment.
"[[DoubleBrackets]]" to generate API doc link
|
Can you resolve merge conflicts? I made a quick pass for style and organization; did not check correctness. Overall it looks like there is quite a bit of repeated code between here and |
added 7 commits
October 5, 2015 15:40
o ProbabilisticClassifier.scala:
mention calibration in comments
o BinaryClassificationMetrics.scala:
adapting code for ROC to calibration; incomplete and commented
out for now
o BinaryClassificationMetricsSuite.scala:
tests for calibration
types to what calibration actually returns.
same as spark master again.
|
I'm going to close this pull request. If this is still relevant and you are interested in pushing it forward, please open a new pull request. Thanks! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR contains an implementation of a calibration method in the class BinaryClassificationMetrics. The code was adapted from the method for ROC curve construction. Tests on small data sets have been added to BinaryClassificationMetricsSuite, and the current version of the code passes those tests.
In this implementation, the return value of the new method is an RDD[((Double, Double), (Double, Long))]. The first pair describes each bin and the second pair describes the content of each bin. In the first pair, the two values are the least and greatest scores in the bin. In the second pair, the two values are the proportion of positive examples in the bin, and the number of examples in the bin. I chose this representation in order to keep as much information as possible. However, a simpler representation might be better; let's talk about that if anyone is interested.