You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@spark.apache.org by we...@apache.org on 2017/09/21 13:43:21 UTC
[4/4] spark git commit: [SPARK-17997][SQL] Add an aggregation
function for counting distinct values for multiple intervals
[SPARK-17997][SQL] Add an aggregation function for counting distinct values for multiple intervals
## What changes were proposed in this pull request?
This work is a part of [SPARK-17074](https://issues.apache.org/jira/browse/SPARK-17074) to compute equi-height histograms. Equi-height histogram is an array of bins. A bin consists of two endpoints which form an interval of values and the ndv in that interval.
This PR creates a new aggregate function, given an array of endpoints, counting distinct values (ndv) in intervals among those endpoints.
This PR also refactors `HyperLogLogPlusPlus` by extracting a helper class `HyperLogLogPlusPlusHelper`, where the underlying HLLPP algorithm locates.
## How was this patch tested?
Add new test cases.
Author: Zhenhua Wang <wa...@huawei.com>
Closes #15544 from wzhfy/countIntervals.
Project: http://git-wip-us.apache.org/repos/asf/spark/repo
Commit: http://git-wip-us.apache.org/repos/asf/spark/commit/1d1a09be
Tree: http://git-wip-us.apache.org/repos/asf/spark/tree/1d1a09be
Diff: http://git-wip-us.apache.org/repos/asf/spark/diff/1d1a09be
Branch: refs/heads/master
Commit: 1d1a09be9fe07244d7d12442f6105f823b260755
Parents: a8d9ec8
Author: Zhenhua Wang <wa...@huawei.com>
Authored: Thu Sep 21 21:43:02 2017 +0800
Committer: Wenchen Fan <we...@databricks.com>
Committed: Thu Sep 21 21:43:02 2017 +0800
----------------------------------------------------------------------
.../ApproxCountDistinctForIntervals.scala | 242 ++++++++++++
.../aggregate/HyperLogLogPlusPlus.scala | 346 +----------------
.../util/HyperLogLogPlusPlusHelper.scala | 373 +++++++++++++++++++
.../ApproxCountDistinctForIntervalsSuite.scala | 238 ++++++++++++
.../aggregate/HyperLogLogPlusPlusSuite.scala | 4 +-
5 files changed, 869 insertions(+), 334 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/spark/blob/1d1a09be/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/aggregate/ApproxCountDistinctForIntervals.scala
----------------------------------------------------------------------
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/aggregate/ApproxCountDistinctForIntervals.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/aggregate/ApproxCountDistinctForIntervals.scala
new file mode 100644
index 0000000..096d1b3
--- /dev/null
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/aggregate/ApproxCountDistinctForIntervals.scala
@@ -0,0 +1,242 @@
+/*
+ * 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.sql.catalyst.expressions.aggregate
+
+import java.util
+
+import org.apache.spark.sql.catalyst.InternalRow
+import org.apache.spark.sql.catalyst.analysis.TypeCheckResult
+import org.apache.spark.sql.catalyst.analysis.TypeCheckResult.{TypeCheckFailure, TypeCheckSuccess}
+import org.apache.spark.sql.catalyst.expressions.{AttributeReference, ExpectsInputTypes, Expression}
+import org.apache.spark.sql.catalyst.util.{ArrayData, GenericArrayData, HyperLogLogPlusPlusHelper}
+import org.apache.spark.sql.types._
+
+/**
+ * This function counts the approximate number of distinct values (ndv) in
+ * intervals constructed from endpoints specified in `endpointsExpression`. The endpoints should be
+ * sorted into ascending order. E.g., given an array of endpoints
+ * (endpoint_1, endpoint_2, ... endpoint_N), returns the approximate ndv's for intervals
+ * [endpoint_1, endpoint_2], (endpoint_2, endpoint_3], ... (endpoint_N-1, endpoint_N].
+ * To count ndv's in these intervals, apply the HyperLogLogPlusPlus algorithm in each of them.
+ * @param child to estimate the ndv's of.
+ * @param endpointsExpression An array expression to construct the intervals. It must be foldable,
+ * and its elements should be sorted into ascending order.
+ * Duplicate endpoints are allowed, e.g. (1, 5, 5, 10), and ndv for
+ * interval (5, 5] would be 1.
+ * @param relativeSD The maximum estimation error allowed in the HyperLogLogPlusPlus algorithm.
+ */
+case class ApproxCountDistinctForIntervals(
+ child: Expression,
+ endpointsExpression: Expression,
+ relativeSD: Double = 0.05,
+ mutableAggBufferOffset: Int = 0,
+ inputAggBufferOffset: Int = 0)
+ extends ImperativeAggregate with ExpectsInputTypes {
+
+ def this(child: Expression, endpointsExpression: Expression) = {
+ this(
+ child = child,
+ endpointsExpression = endpointsExpression,
+ relativeSD = 0.05,
+ mutableAggBufferOffset = 0,
+ inputAggBufferOffset = 0)
+ }
+
+ def this(child: Expression, endpointsExpression: Expression, relativeSD: Expression) = {
+ this(
+ child = child,
+ endpointsExpression = endpointsExpression,
+ relativeSD = HyperLogLogPlusPlus.validateDoubleLiteral(relativeSD),
+ mutableAggBufferOffset = 0,
+ inputAggBufferOffset = 0)
+ }
+
+ override def inputTypes: Seq[AbstractDataType] = {
+ Seq(TypeCollection(NumericType, TimestampType, DateType), ArrayType)
+ }
+
+ // Mark as lazy so that endpointsExpression is not evaluated during tree transformation.
+ lazy val endpoints: Array[Double] =
+ (endpointsExpression.dataType, endpointsExpression.eval()) match {
+ case (ArrayType(elementType, _), arrayData: ArrayData) =>
+ arrayData.toObjectArray(elementType).map(_.toString.toDouble)
+ }
+
+ override def checkInputDataTypes(): TypeCheckResult = {
+ val defaultCheck = super.checkInputDataTypes()
+ if (defaultCheck.isFailure) {
+ defaultCheck
+ } else if (!endpointsExpression.foldable) {
+ TypeCheckFailure("The endpoints provided must be constant literals")
+ } else {
+ endpointsExpression.dataType match {
+ case ArrayType(_: NumericType | DateType | TimestampType, _) =>
+ if (endpoints.length < 2) {
+ TypeCheckFailure("The number of endpoints must be >= 2 to construct intervals")
+ } else {
+ TypeCheckSuccess
+ }
+ case _ =>
+ TypeCheckFailure("Endpoints require (numeric or timestamp or date) type")
+ }
+ }
+ }
+
+ // N endpoints construct N-1 intervals, creating a HLLPP for each interval
+ private lazy val hllppArray = {
+ val array = new Array[HyperLogLogPlusPlusHelper](endpoints.length - 1)
+ for (i <- array.indices) {
+ array(i) = new HyperLogLogPlusPlusHelper(relativeSD)
+ }
+ // `numWords` in each HLLPPHelper should be the same because it is determined by `relativeSD`
+ // which is shared among all HLLPPHelpers.
+ assert(array.map(_.numWords).distinct.length == 1)
+ array
+ }
+
+ private lazy val numWordsPerHllpp = hllppArray.head.numWords
+
+ private lazy val totalNumWords = numWordsPerHllpp * hllppArray.length
+
+ /** Allocate enough words to store all registers. */
+ override lazy val aggBufferAttributes: Seq[AttributeReference] = {
+ Seq.tabulate(totalNumWords) { i =>
+ AttributeReference(s"MS[$i]", LongType)()
+ }
+ }
+
+ override def aggBufferSchema: StructType = StructType.fromAttributes(aggBufferAttributes)
+
+ // Note: although this simply copies aggBufferAttributes, this common code can not be placed
+ // in the superclass because that will lead to initialization ordering issues.
+ override lazy val inputAggBufferAttributes: Seq[AttributeReference] =
+ aggBufferAttributes.map(_.newInstance())
+
+ /** Fill all words with zeros. */
+ override def initialize(buffer: InternalRow): Unit = {
+ var word = 0
+ while (word < totalNumWords) {
+ buffer.setLong(mutableAggBufferOffset + word, 0)
+ word += 1
+ }
+ }
+
+ override def update(buffer: InternalRow, input: InternalRow): Unit = {
+ val value = child.eval(input)
+ // Ignore empty rows
+ if (value != null) {
+ // convert the value into a double value for searching in the double array
+ val doubleValue = child.dataType match {
+ case n: NumericType =>
+ n.numeric.toDouble(value.asInstanceOf[n.InternalType])
+ case _: DateType =>
+ value.asInstanceOf[Int].toDouble
+ case _: TimestampType =>
+ value.asInstanceOf[Long].toDouble
+ }
+
+ // endpoints are sorted into ascending order already
+ if (endpoints.head > doubleValue || endpoints.last < doubleValue) {
+ // ignore if the value is out of the whole range
+ return
+ }
+
+ val hllppIndex = findHllppIndex(doubleValue)
+ val offset = mutableAggBufferOffset + hllppIndex * numWordsPerHllpp
+ hllppArray(hllppIndex).update(buffer, offset, value, child.dataType)
+ }
+ }
+
+ // Find which interval (HyperLogLogPlusPlusHelper) should receive the given value.
+ def findHllppIndex(value: Double): Int = {
+ var index = util.Arrays.binarySearch(endpoints, value)
+ if (index >= 0) {
+ // The value is found.
+ if (index == 0) {
+ 0
+ } else {
+ // If the endpoints contains multiple elements with the specified value, there is no
+ // guarantee which one binarySearch will return. We remove this uncertainty by moving the
+ // index to the first position of these elements.
+ var first = index - 1
+ while (first >= 0 && endpoints(first) == value) {
+ first -= 1
+ }
+ index = first + 1
+
+ if (index == 0) {
+ // reach the first endpoint
+ 0
+ } else {
+ // send values in (endpoints(index-1), endpoints(index)] to hllpps(index-1)
+ index - 1
+ }
+ }
+ } else {
+ // The value is not found, binarySearch returns (-(<i>insertion point</i>) - 1).
+ // The <i>insertion point</i> is defined as the point at which the key would be inserted
+ // into the array: the index of the first element greater than the key.
+ val insertionPoint = - (index + 1)
+ if (insertionPoint == 0) 0 else insertionPoint - 1
+ }
+ }
+
+ override def merge(buffer1: InternalRow, buffer2: InternalRow): Unit = {
+ for (i <- hllppArray.indices) {
+ hllppArray(i).merge(
+ buffer1 = buffer1,
+ buffer2 = buffer2,
+ offset1 = mutableAggBufferOffset + i * numWordsPerHllpp,
+ offset2 = inputAggBufferOffset + i * numWordsPerHllpp)
+ }
+ }
+
+ override def eval(buffer: InternalRow): Any = {
+ val ndvArray = hllppResults(buffer)
+ // If the endpoints contains multiple elements with the same value,
+ // we set ndv=1 for intervals between these elements.
+ // E.g. given four endpoints (1, 2, 2, 4) and input sequence (0.5, 2),
+ // the ndv's for the three intervals should be (2, 1, 0)
+ for (i <- ndvArray.indices) {
+ if (endpoints(i) == endpoints(i + 1)) ndvArray(i) = 1
+ }
+ new GenericArrayData(ndvArray)
+ }
+
+ def hllppResults(buffer: InternalRow): Array[Long] = {
+ val ndvArray = new Array[Long](hllppArray.length)
+ for (i <- ndvArray.indices) {
+ ndvArray(i) = hllppArray(i).query(buffer, mutableAggBufferOffset + i * numWordsPerHllpp)
+ }
+ ndvArray
+ }
+
+ override def withNewMutableAggBufferOffset(newMutableAggBufferOffset: Int): ImperativeAggregate =
+ copy(mutableAggBufferOffset = newMutableAggBufferOffset)
+
+ override def withNewInputAggBufferOffset(newInputAggBufferOffset: Int): ImperativeAggregate =
+ copy(inputAggBufferOffset = newInputAggBufferOffset)
+
+ override def children: Seq[Expression] = Seq(child, endpointsExpression)
+
+ override def nullable: Boolean = false
+
+ override def dataType: DataType = ArrayType(LongType)
+
+ override def prettyName: String = "approx_count_distinct_for_intervals"
+}
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@spark.apache.org
For additional commands, e-mail: commits-help@spark.apache.org