Github user cloud-fan commented on a diff in the pull request:
https://github.com/apache/spark/pull/14298#discussion_r73832267
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/aggregate/PercentileApprox.scala
---
@@ -0,0 +1,462 @@
+/*
+ * 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 scala.collection.mutable.ArrayBuffer
+
+import org.apache.spark.sql.AnalysisException
+import org.apache.spark.sql.catalyst.InternalRow
+import org.apache.spark.sql.catalyst.analysis.TypeCheckResult
+import org.apache.spark.sql.catalyst.expressions._
+import
org.apache.spark.sql.catalyst.expressions.aggregate.QuantileSummaries.Stats
+import org.apache.spark.sql.catalyst.util._
+import org.apache.spark.sql.types._
+
+/**
+ * Computes an approximate percentile (quantile) using the G-K algorithm
(see below), for very
+ * large numbers of rows where the regular percentile() UDAF might run out
of memory.
+ *
+ * The input is a single double value or an array of double values
representing the percentiles
+ * requested. The output, corresponding to the input, is either a single
double value or an
+ * array of doubles that are the percentile values.
+ */
+@ExpressionDescription(
+ usage = """_FUNC_(col, p [, B]) - Returns an approximate pth percentile
of a numeric column in the
+ group. The B parameter, which defaults to 1000, controls
approximation accuracy at the cost of
+ memory; higher values yield better approximations.
+ _FUNC_(col, array(p1 [, p2]...) [, B]) - Same as above, but accepts
and returns an array of
+ percentile values instead of a single one.
+ """)
+case class PercentileApprox(
+ child: Expression,
+ percentilesExpr: Expression,
+ bExpr: Option[Expression],
+ percentiles: Seq[Double], // the extracted percentiles
+ B: Int, // the extracted B
+ resultAsArray: Boolean, // whether to return the result as an array
+ mutableAggBufferOffset: Int = 0,
+ inputAggBufferOffset: Int = 0) extends ImperativeAggregate {
+
+ private def this(child: Expression, percentilesExpr: Expression, bExpr:
Option[Expression]) = {
+ this(
+ child = child,
+ percentilesExpr = percentilesExpr,
+ bExpr = bExpr,
+ // validate and extract percentiles
+ percentiles =
PercentileApprox.validatePercentilesLiteral(percentilesExpr)._1,
+ // validate and extract B
+ B =
bExpr.map(PercentileApprox.validateBLiteral(_)).getOrElse(PercentileApprox.B_DEFAULT),
+ // validate and mark whether we should return results as array of
double or not
+ resultAsArray =
PercentileApprox.validatePercentilesLiteral(percentilesExpr)._2)
+ }
+
+ // Constructor for the "_FUNC_(col, p) / _FUNC_(col, array(p1, ...))"
form
+ def this(child: Expression, percentilesExpr: Expression) = {
+ this(child, percentilesExpr, None)
+ }
+
+ // Constructor for the "_FUNC_(col, p, B) / _FUNC_(col, array(p1, ...),
B)" form
+ def this(child: Expression, percentilesExpr: Expression, bExpr:
Expression) = {
+ this(child, percentilesExpr, Some(bExpr))
+ }
+
+ override def prettyName: String = "percentile_approx"
+
+ override def withNewMutableAggBufferOffset(newMutableAggBufferOffset:
Int): ImperativeAggregate =
+ copy(mutableAggBufferOffset = newMutableAggBufferOffset)
+
+ override def withNewInputAggBufferOffset(newInputAggBufferOffset: Int):
ImperativeAggregate =
+ copy(inputAggBufferOffset = newInputAggBufferOffset)
+
+ override def children: Seq[Expression] =
+ bExpr.map(child :: percentilesExpr :: _ :: Nil).getOrElse(child ::
percentilesExpr :: Nil)
+
+ // we would return null for empty inputs
+ override def nullable: Boolean = true
+
+ override def dataType: DataType = if (resultAsArray)
ArrayType(DoubleType) else DoubleType
+
+ override def inputTypes: Seq[AbstractDataType] = Seq(NumericType,
AnyDataType, IntegralType)
+
+ override def checkInputDataTypes(): TypeCheckResult =
+ TypeUtils.checkForNumericExpr(child.dataType, "function
percentile_approx")
+
+ // The number of intermediate outputs is highly relative to the actual
data-set (an upper bound is
+ // (11/2e)log(2en), where e is the relativeError parameter, n is the
number of items in the
+ // dataset) -- thus it's hard to allocate agg buffer in advance without
knowing the size of
+ // inputs. Due to this reason, currently we don't support partial mode.
--- End diff --
can you explain a bit more about this? AFAIK, hive supports partial
aggregate for `percentile_approx`, and it looks to me that your implementation
keeps the buffer data(`QuantileSummaries`) in this aggregate function object,
instead of letting aggregate operator manage it, that's the main reason why we
can't support partial aggregate for `percentile_approx` I think.
---
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 infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
