org.apache.spark.partial.SumEvaluator.scala Maven / Gradle / Ivy
/*
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.spark.partial
import org.apache.commons.math3.distribution.{NormalDistribution, TDistribution}
import org.apache.spark.util.StatCounter
/**
* An ApproximateEvaluator for sums. It estimates the mean and the count and multiplies them
* together, then uses the formula for the variance of two independent random variables to get
* a variance for the result and compute a confidence interval.
*/
private[spark] class SumEvaluator(totalOutputs: Int, confidence: Double)
extends ApproximateEvaluator[StatCounter, BoundedDouble] {
// modified in merge
private var outputsMerged = 0
private val counter = new StatCounter()
override def merge(outputId: Int, taskResult: StatCounter): Unit = {
outputsMerged += 1
counter.merge(taskResult)
}
override def currentResult(): BoundedDouble = {
if (outputsMerged == totalOutputs) {
new BoundedDouble(counter.sum, 1.0, counter.sum, counter.sum)
} else if (outputsMerged == 0 || counter.count == 0) {
new BoundedDouble(0, 0.0, Double.NegativeInfinity, Double.PositiveInfinity)
} else {
val p = outputsMerged.toDouble / totalOutputs
// Expected value of unobserved is presumed equal to that of the observed data
val meanEstimate = counter.mean
// Expected size of rest of the data is proportional
val countEstimate = counter.count * (1 - p) / p
// Expected sum is simply their product
val sumEstimate = meanEstimate * countEstimate
// Variance of unobserved data is presumed equal to that of the observed data
val meanVar = counter.sampleVariance / counter.count
// branch at this point because count == 1 implies counter.sampleVariance == Nan
// and we don't want to ever return a bound of NaN
if (meanVar.isNaN || counter.count == 1) {
// add sum because estimate is of unobserved data sum
new BoundedDouble(
counter.sum + sumEstimate, confidence, Double.NegativeInfinity, Double.PositiveInfinity)
} else {
// See CountEvaluator. Variance of population count here follows from negative binomial
val countVar = counter.count * (1 - p) / (p * p)
// Var(Sum) = Var(Mean*Count) =
// [E(Mean)]^2 * Var(Count) + [E(Count)]^2 * Var(Mean) + Var(Mean) * Var(Count)
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 + confidence) / 2)
} else {
// note that if this goes to 0, TDistribution will throw an exception.
// Hence special casing 1 above.
val degreesOfFreedom = (counter.count - 1).toInt
new TDistribution(degreesOfFreedom).inverseCumulativeProbability((1 + confidence) / 2)
}
// Symmetric, so confidence interval is symmetric about mean of distribution
val low = sumEstimate - confFactor * sumStdev
val high = sumEstimate + confFactor * sumStdev
// add sum because estimate is of unobserved data sum
new BoundedDouble(
counter.sum + sumEstimate, confidence, counter.sum + low, counter.sum + high)
}
}
}
}