[SPARK-4256] Make Binary Evaluation Metrics functions defined in cases where there ar... #3118
Conversation
… are 0 positive or 0 negative examples.
|
Can one of the admins verify this patch? |
| @@ -27,19 +27,31 @@ private[evaluation] trait BinaryClassificationMetricComputer extends Serializabl | |||
| /** Precision. */ | |||
| private[evaluation] object Precision extends BinaryClassificationMetricComputer { | |||
| override def apply(c: BinaryConfusionMatrix): Double = | |||
| c.numTruePositives.toDouble / (c.numTruePositives + c.numFalsePositives) | |||
| if (c.numTruePositives + c.numFalsePositives == 0) { | |||
There was a problem hiding this comment.
Use another variable for brevity?
val totalPositives = c.numTruePositives + c.numFalsePositives
|
@abull Thanks for noticing this end case! I made some small inline comments. One more: Since end-cases are not well-defined for these types of methods, could you please add comments in the methods' docs about behavior when the denominator is 0? |
|
Can one of the admins verify this patch? |
|
Sure, will update the PR. (Hopefully some time today) |
…les; update code formatting per pull request comments
…on functions in test
|
Cool, I updated the code per the feedback above. |
| } | ||
|
|
||
| private def validateMetrics(metrics: BinaryClassificationMetrics, | ||
| expectedThresholds: Seq[Double], |
There was a problem hiding this comment.
Please use 4-space indentation. We don't use vertical alignment in Spark.
|
Updated the test code formatting per comments from @mengxr |
|
ok to test |
|
test this please |
|
LGTM. Thank you for checking the corner cases! Waiting for Jenkins ... |
|
Test build #23292 has started for PR 3118 at commit
|
|
Test build #23292 has finished for PR 3118 at commit
|
|
Test PASSed. |
|
Merged into master and branch-1.2. Thanks! |
…s where there ar... ...e 0 positive or 0 negative examples. Author: Andrew Bullen <andrew.bullen@workday.com> Closes #3118 from abull/master and squashes the following commits: c2bf2b1 [Andrew Bullen] [SPARK-4256] Update Code formatting for BinaryClassificationMetricsSpec 36b0533 [Andrew Bullen] [SYMAN-4256] Extract BinaryClassificationMetricsSuite assertions into private method 4d2f79a [Andrew Bullen] [SPARK-4256] Refactor classification metrics tests - extract comparison functions in test f411e70 [Andrew Bullen] [SPARK-4256] Define precision as 1.0 when there are no positive examples; update code formatting per pull request comments d9a09ef [Andrew Bullen] Make Binary Evaluation Metrics functions defined in cases where there are 0 positive or 0 negative examples. (cherry picked from commit 484fecb) Signed-off-by: Xiangrui Meng <meng@databricks.com>
...e 0 positive or 0 negative examples.