Github user srowen commented on a diff in the pull request:
https://github.com/apache/spark/pull/12016#discussion_r57732511
--- Diff: core/src/main/scala/org/apache/spark/partial/SumEvaluator.scala
---
@@ -40,30 +41,39 @@ private[spark] class SumEvaluator(totalOutputs: Int,
confidence: Double)
override def currentResult(): BoundedDouble = {
if (outputsMerged == totalOutputs) {
new BoundedDouble(counter.sum, 1.0, counter.sum, counter.sum)
- } else if (outputsMerged == 0) {
+ } else if (outputsMerged == 0 || counter.count == 0) {
new BoundedDouble(0, 0.0, Double.NegativeInfinity,
Double.PositiveInfinity)
} else {
val p = outputsMerged.toDouble / totalOutputs
val meanEstimate = counter.mean
- val meanVar = counter.sampleVariance / counter.count
val countEstimate = (counter.count + 1 - p) / p
- val countVar = (counter.count + 1) * (1 - p) / (p * p)
val sumEstimate = meanEstimate * countEstimate
- val sumVar = (meanEstimate * meanEstimate * countVar) +
- (countEstimate * countEstimate * meanVar) +
- (meanVar * countVar)
- val sumStdev = math.sqrt(sumVar)
- val confFactor = {
- if (counter.count > 100) {
+
+ val meanVar = counter.sampleVariance / counter.count
+
+ // branch at this point because counter.count == 1 implies
counter.sampleVariance == Nan
+ // and we don't want to ever return a bound of NaN
+ if (meanVar == Double.NaN || counter.count == 1) {
+ new BoundedDouble(sumEstimate, confidence,
Double.NegativeInfinity, Double.PositiveInfinity)
+ } else {
+ val countVar = (counter.count + 1) * (1 - p) / (p * p)
+ val sumVar = (meanEstimate * meanEstimate * countVar) +
+ (countEstimate * countEstimate * meanVar) +
+ (meanVar * countVar)
+ val sumStdev = math.sqrt(sumVar)
+ val confFactor = if (counter.count > 100) {
new NormalDistribution().inverseCumulativeProbability(1 - (1 -
confidence) / 2)
- } else {
+ } else if (counter.count > 1) {
val degreesOfFreedom = (counter.count - 1).toInt
new
TDistribution(degreesOfFreedom).inverseCumulativeProbability(1 - (1 -
confidence) / 2)
+ } else {
+ throw new Exception("Counter.count <= 1; this should be
impossible at this point")
--- End diff --
You've already handled the count=0 and count=1 cases earlier. Checking
count > 1 doesn't do anything since it can't happen so having a branch for it
is odd. Tests are how we catch regressions.
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
