Github user yanboliang commented on a diff in the pull request:
https://github.com/apache/spark/pull/18798#discussion_r133111870
--- Diff: mllib/src/main/scala/org/apache/spark/ml/stat/Summarizer.scala ---
@@ -0,0 +1,593 @@
+/*
+ * 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.ml.stat
+
+import java.io._
+
+import org.apache.spark.annotation.Since
+import org.apache.spark.internal.Logging
+import org.apache.spark.ml.linalg.{Vector, Vectors, VectorUDT}
+import org.apache.spark.sql.Column
+import org.apache.spark.sql.catalyst.InternalRow
+import org.apache.spark.sql.catalyst.expressions.{Expression,
UnsafeArrayData}
+import
org.apache.spark.sql.catalyst.expressions.aggregate.{AggregateExpression,
Complete, TypedImperativeAggregate}
+import org.apache.spark.sql.catalyst.util.ArrayData
+import org.apache.spark.sql.functions.lit
+import org.apache.spark.sql.types._
+
+/**
+ * A builder object that provides summary statistics about a given column.
+ *
+ * Users should not directly create such builders, but instead use one of
the methods in
+ * [[Summarizer]].
+ */
+@Since("2.3.0")
+abstract class SummaryBuilder {
+ /**
+ * Returns an aggregate object that contains the summary of the column
with the requested metrics.
+ * @param featuresCol a column that contains features Vector object.
+ * @param weightCol a column that contains weight value.
+ * @return an aggregate column that contains the statistics. The exact
content of this
+ * structure is determined during the creation of the builder.
+ */
+ @Since("2.3.0")
+ def summary(featuresCol: Column, weightCol: Column): Column
+
+ @Since("2.3.0")
+ def summary(featuresCol: Column): Column = summary(featuresCol, lit(1.0))
+}
+
+/**
+ * Tools for vectorized statistics on MLlib Vectors.
+ *
+ * The methods in this package provide various statistics for Vectors
contained inside DataFrames.
+ *
+ * This class lets users pick the statistics they would like to extract
for a given column. Here is
+ * an example in Scala:
+ * {{{
+ * val dataframe = ... // Some dataframe containing a feature column
+ * val allStats = dataframe.select(Summarizer.metrics("min",
"max").summary($"features"))
+ * val Row(min_, max_) = allStats.first()
+ * }}}
+ *
+ * If one wants to get a single metric, shortcuts are also available:
+ * {{{
+ * val meanDF = dataframe.select(Summarizer.mean($"features"))
+ * val Row(mean_) = meanDF.first()
+ * }}}
+ *
+ * Note: Currently, the performance of this interface is about 2x~3x
slower then using the RDD
+ * interface.
+ */
+@Since("2.3.0")
+object Summarizer extends Logging {
+
+ import SummaryBuilderImpl._
+
+ /**
+ * Given a list of metrics, provides a builder that it turns computes
metrics from a column.
+ *
+ * See the documentation of [[Summarizer]] for an example.
+ *
+ * The following metrics are accepted (case sensitive):
+ * - mean: a vector that contains the coefficient-wise mean.
+ * - variance: a vector tha contains the coefficient-wise variance.
+ * - count: the count of all vectors seen.
+ * - numNonzeros: a vector with the number of non-zeros for each
coefficients
+ * - max: the maximum for each coefficient.
+ * - min: the minimum for each coefficient.
+ * - normL2: the Euclidian norm for each coefficient.
+ * - normL1: the L1 norm of each coefficient (sum of the absolute
values).
+ * @param firstMetric the metric being provided
+ * @param metrics additional metrics that can be provided.
+ * @return a builder.
+ * @throws IllegalArgumentException if one of the metric names is not
understood.
+ *
+ * Note: Currently, the performance of this interface is about 2x~3x
slower then using the RDD
+ * interface.
+ */
+ @Since("2.3.0")
+ def metrics(firstMetric: String, metrics: String*): SummaryBuilder = {
+ val (typedMetrics, computeMetrics) =
getRelevantMetrics(Seq(firstMetric) ++ metrics)
+ new SummaryBuilderImpl(typedMetrics, computeMetrics)
+ }
+
+ def mean(col: Column): Column = getSingleMetric(col, "mean")
+
+ def variance(col: Column): Column = getSingleMetric(col, "variance")
+
+ def count(col: Column): Column = getSingleMetric(col, "count")
+
+ def numNonZeros(col: Column): Column = getSingleMetric(col,
"numNonZeros")
+
+ def max(col: Column): Column = getSingleMetric(col, "max")
+
+ def min(col: Column): Column = getSingleMetric(col, "min")
+
+ def normL1(col: Column): Column = getSingleMetric(col, "normL1")
+
+ def normL2(col: Column): Column = getSingleMetric(col, "normL2")
+
+ private def getSingleMetric(col: Column, metric: String): Column = {
+ val c1 = metrics(metric).summary(col)
+ c1.getField(metric).as(s"$metric($col)")
+ }
+}
+
+private[ml] class SummaryBuilderImpl(
+ requestedMetrics: Seq[SummaryBuilderImpl.Metrics],
+ requestedCompMetrics: Seq[SummaryBuilderImpl.ComputeMetrics]
+ ) extends SummaryBuilder {
+
+ override def summary(featuresCol: Column, weightCol: Column): Column = {
+
+ val agg = SummaryBuilderImpl.MetricsAggregate(
+ requestedMetrics,
+ requestedCompMetrics,
+ featuresCol.expr,
+ weightCol.expr,
+ mutableAggBufferOffset = 0,
+ inputAggBufferOffset = 0)
+
+ new Column(AggregateExpression(agg, mode = Complete, isDistinct =
false))
+ }
+}
+
+private[ml]
+object SummaryBuilderImpl extends Logging {
+
+ def implementedMetrics: Seq[String] = allMetrics.map(_._1).sorted
+
+ @throws[IllegalArgumentException]("When the list is empty or not a
subset of known metrics")
+ def getRelevantMetrics(requested: Seq[String]): (Seq[Metrics],
Seq[ComputeMetrics]) = {
+ val all = requested.map { req =>
+ val (_, metric, _, deps) = allMetrics.find(tup => tup._1 ==
req).getOrElse {
+ throw new IllegalArgumentException(s"Metric $req cannot be found."
+
+ s" Valid metrics are $implementedMetrics")
+ }
+ metric -> deps
+ }
+ // Do not sort, otherwise the user has to look the schema to see the
order that it
+ // is going to be given in.
+ val metrics = all.map(_._1)
+ val computeMetrics = all.flatMap(_._2).distinct.sortBy(_.toString)
+ metrics -> computeMetrics
+ }
+
+ def structureForMetrics(metrics: Seq[Metrics]): StructType = {
+ val dct = allMetrics.map { case (n, m, dt, _) => (m, (n, dt)) }.toMap
+ val fields = metrics.map(dct.apply).map { case (n, dt) =>
+ StructField(n, dt, nullable = false)
+ }
+ StructType(fields)
+ }
+
+ private val arrayDType = ArrayType(DoubleType, containsNull = false)
+ private val arrayLType = ArrayType(LongType, containsNull = false)
+
+ /**
+ * All the metrics that can be currently computed by Spark for vectors.
+ *
+ * This list associates the user name, the internal (typed) name, and
the list of computation
+ * metrics that need to de computed internally to get the final result.
+ */
+ private val allMetrics: Seq[(String, Metrics, DataType,
Seq[ComputeMetrics])] = Seq(
+ ("mean", Mean, arrayDType, Seq(ComputeMean, ComputeWeightSum)),
+ ("variance", Variance, arrayDType, Seq(ComputeWeightSum, ComputeMean,
ComputeM2n)),
+ ("count", Count, LongType, Seq()),
+ ("numNonZeros", NumNonZeros, arrayLType, Seq(ComputeNNZ)),
+ ("max", Max, arrayDType, Seq(ComputeMax, ComputeNNZ)),
+ ("min", Min, arrayDType, Seq(ComputeMin, ComputeNNZ)),
+ ("normL2", NormL2, arrayDType, Seq(ComputeM2)),
+ ("normL1", NormL1, arrayDType, Seq(ComputeL1))
+ )
+
+ /**
+ * The metrics that are currently implemented.
+ */
+ sealed trait Metrics extends Serializable
+ case object Mean extends Metrics
--- End diff --
Should we keep these case objects private?
---
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
