Github user ebernhardson commented on a diff in the pull request:
https://github.com/apache/spark/pull/16618#discussion_r113355473
--- Diff:
mllib/src/main/scala/org/apache/spark/ml/evaluation/RankingMetrics.scala ---
@@ -0,0 +1,202 @@
+/*
+ * 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.evaluation
+
+import org.apache.spark.annotation.Since
+import org.apache.spark.internal.Logging
+import org.apache.spark.sql.{Column, DataFrame}
+import org.apache.spark.sql.functions.{mean, sum}
+import org.apache.spark.sql.functions.udf
+import org.apache.spark.sql.types.DoubleType
+
+@Since("2.2.0")
+class RankingMetrics(
+ predictionAndObservations: DataFrame, predictionCol: String, labelCol:
String)
+ extends Logging with Serializable {
+
+ /**
+ * Compute the Mean Percentile Rank (MPR) of all the queries.
+ *
+ * See the following paper for detail ("Expected percentile rank" in the
paper):
+ * Hu, Y., Y. Koren, and C. Volinsky. âCollaborative Filtering for
Implicit Feedback Datasets.â
+ * In 2008 Eighth IEEE International Conference on Data Mining,
263â72, 2008.
+ * doi:10.1109/ICDM.2008.22.
+ *
+ * @return the mean percentile rank
+ */
+ lazy val meanPercentileRank: Double = {
+
+ def rank = udf((predicted: Seq[Any], actual: Any) => {
+ val l_i = predicted.indexOf(actual)
+
+ if (l_i == -1) {
+ 1
+ } else {
+ l_i.toDouble / predicted.size
+ }
+ }, DoubleType)
+
+ val R_prime = predictionAndObservations.count()
+ val predictionColumn: Column =
predictionAndObservations.col(predictionCol)
+ val labelColumn: Column = predictionAndObservations.col(labelCol)
+
+ val rankSum: Double = predictionAndObservations
+ .withColumn("rank", rank(predictionColumn, labelColumn))
+ .agg(sum("rank")).first().getDouble(0)
+
+ rankSum / R_prime
+ }
+
+ /**
+ * Compute the average precision of all the queries, truncated at
ranking position k.
+ *
+ * If for a query, the ranking algorithm returns n (n is less than k)
results, the precision
+ * value will be computed as #(relevant items retrieved) / k. This
formula also applies when
+ * the size of the ground truth set is less than k.
+ *
+ * If a query has an empty ground truth set, zero will be used as
precision together with
+ * a log warning.
+ *
+ * See the following paper for detail:
+ *
+ * IR evaluation methods for retrieving highly relevant documents. K.
Jarvelin and J. Kekalainen
+ *
+ * @param k the position to compute the truncated precision, must be
positive
+ * @return the average precision at the first k ranking positions
+ */
+ @Since("2.2.0")
+ def precisionAt(k: Int): Double = {
+ require(k > 0, "ranking position k should be positive")
+
+ def precisionAtK = udf((predicted: Seq[Any], actual: Seq[Any]) => {
+ val actualSet = actual.toSet
+ if (actualSet.nonEmpty) {
+ val n = math.min(predicted.length, k)
+ var i = 0
+ var cnt = 0
+ while (i < n) {
+ if (actualSet.contains(predicted(i))) {
+ cnt += 1
+ }
+ i += 1
+ }
+ cnt.toDouble / k
+ } else {
+ logWarning("Empty ground truth set, check input data")
+ 0.0
+ }
+ }, DoubleType)
+
+ val predictionColumn: Column =
predictionAndObservations.col(predictionCol)
+ val labelColumn: Column = predictionAndObservations.col(labelCol)
+
+ predictionAndObservations
+ .withColumn("predictionAtK", precisionAtK(predictionColumn,
labelColumn))
+ .agg(mean("predictionAtK")).first().getDouble(0)
+ }
+
+ /**
+ * Returns the mean average precision (MAP) of all the queries.
+ * If a query has an empty ground truth set, the average precision will
be zero and a log
+ * warning is generated.
+ */
+ lazy val meanAveragePrecision: Double = {
+
+ def map = udf((predicted: Seq[Any], actual: Seq[Any]) => {
+ val actualSet = actual.toSet
+ if (actualSet.nonEmpty) {
+ var i = 0
+ var cnt = 0
+ var precSum = 0.0
+ val n = predicted.length
+ while (i < n) {
+ if (actualSet.contains(predicted(i))) {
+ cnt += 1
+ precSum += cnt.toDouble / (i + 1)
+ }
+ i += 1
+ }
+ precSum / actualSet.size
+ } else {
+ logWarning("Empty ground truth set, check input data")
+ 0.0
+ }
+ }, DoubleType)
+
+ val predictionColumn: Column =
predictionAndObservations.col(predictionCol)
+ val labelColumn: Column = predictionAndObservations.col(labelCol)
+
+ predictionAndObservations
+ .withColumn("MAP", map(predictionColumn, labelColumn))
+ .agg(mean("MAP")).first().getDouble(0)
+ }
+
+ /**
+ * Compute the average NDCG value of all the queries, truncated at
ranking position k.
+ * The discounted cumulative gain at position k is computed as:
+ * sum,,i=1,,^k^ (2^{relevance of ''i''th item}^ - 1) / log(i + 1),
+ * and the NDCG is obtained by dividing the DCG value on the ground
truth set. In the current
+ * implementation, the relevance value is binary.
+
+ * If a query has an empty ground truth set, zero will be used as ndcg
together with
+ * a log warning.
+ *
+ * See the following paper for detail:
+ *
+ * IR evaluation methods for retrieving highly relevant documents. K.
Jarvelin and J. Kekalainen
+ *
+ * @param k the position to compute the truncated ndcg, must be positive
+ * @return the average ndcg at the first k ranking positions
+ */
+ @Since("2.2.0")
+ def ndcgAt(k: Int): Double = {
+ require(k > 0, "ranking position k should be positive")
+
+ def ndcgAtK = udf((predicted: Seq[Any], actual: Seq[Any]) => {
+ val actualSet = actual.toSet
+
+ if (actualSet.nonEmpty) {
+ val labSetSize = actualSet.size
+ val n = math.min(math.max(predicted.length, labSetSize), k)
+ var maxDcg = 0.0
+ var dcg = 0.0
+ var i = 0
+ while (i < n) {
+ val gain = 1.0 / math.log(i + 2)
--- End diff --
Any particular reason to use 1.0 here instead of the label, as described in
the method docs? It might require some re-jiggering as you are accepting
Seq[Any] and you instead need a number of some sort (both ints and floats are
used in practice)
---
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
