Github user MLnick commented on a diff in the pull request:
https://github.com/apache/spark/pull/15148#discussion_r80470751
--- Diff: mllib/src/main/scala/org/apache/spark/ml/feature/lsh/LSH.scala ---
@@ -0,0 +1,304 @@
+/*
+ * 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.feature.lsh
+
+import scala.util.Random
+
+import org.apache.spark.ml.{Estimator, Model}
+import org.apache.spark.ml.linalg.{Vector, VectorUDT}
+import org.apache.spark.ml.param.{IntParam, ParamMap, ParamValidators}
+import org.apache.spark.ml.param.shared.{HasInputCol, HasOutputCol}
+import org.apache.spark.sql._
+import org.apache.spark.sql.expressions.UserDefinedFunction
+import org.apache.spark.sql.functions._
+import org.apache.spark.sql.types._
+
+/**
+ * Params for [[LSH]].
+ */
+private[ml] trait LSHParams extends HasInputCol with HasOutputCol {
+ /**
+ * Param for output dimension.
+ *
+ * @group param
+ */
+ final val outputDim: IntParam = new IntParam(this, "outputDim", "output
dimension",
+ ParamValidators.gt(0))
+
+ /** @group getParam */
+ final def getOutputDim: Int = $(outputDim)
+
+ setDefault(outputDim -> 1)
+
+ setDefault(outputCol -> "lsh_output")
+
+ /**
+ * Transform the Schema for LSH
+ * @param schema The schema of the input dataset without outputCol
+ * @return A derived schema with outputCol added
+ */
+ final def transformLSHSchema(schema: StructType): StructType = {
+ val outputFields = schema.fields :+
+ StructField($(outputCol), new VectorUDT, nullable = false)
+ StructType(outputFields)
+ }
+}
+
+/**
+ * Model produced by [[LSH]].
+ */
+abstract class LSHModel[KeyType, T <: LSHModel[KeyType, T]] private[ml]
+ extends Model[T] with LSHParams {
+ override def copy(extra: ParamMap): T = defaultCopy(extra)
+ /**
+ * :: DeveloperApi ::
+ *
+ * The hash function of LSH, mapping a predefined KeyType to a Vector
+ * @return The mapping of LSH function.
+ */
+ protected[this] val hashFunction: KeyType => Vector
+
+ /**
+ * :: DeveloperApi ::
+ *
+ * Calculate the distance between two different keys using the distance
metric corresponding
+ * to the hashFunction
+ * @param x One of the point in the metric space
+ * @param y Another the point in the metric space
+ * @return The distance between x and y in double
+ */
+ protected[ml] def keyDistance(x: KeyType, y: KeyType): Double
+
+ /**
+ * :: DeveloperApi ::
+ *
+ * Calculate the distance between two different hash Vectors. By
default, the distance is the
+ * minimum distance of two hash values in any dimension.
+ *
+ * @param x One of the hash vector
+ * @param y Another hash vector
+ * @return The distance between hash vectors x and y in double
+ */
+ protected[ml] def hashDistance(x: Vector, y: Vector): Double = {
+ // Since it's generated by hashing, it will be a pair of dense vectors.
+ x.toDense.values.zip(y.toDense.values).map(x => math.abs(x._1 -
x._2)).min
+ }
+
+ /**
+ * Transforms the input dataset.
+ */
+ override def transform(dataset: Dataset[_]): DataFrame = {
+ if (!dataset.columns.contains($(outputCol))) {
+ transformSchema(dataset.schema, logging = true)
+ val transformUDF = udf(hashFunction, new VectorUDT)
+ dataset.withColumn($(outputCol), transformUDF(dataset($(inputCol))))
+ } else {
+ dataset.toDF()
+ }
+ }
+
+ /**
+ * :: DeveloperApi ::
+ *
+ * Check transform validity and derive the output schema from the input
schema.
+ *
+ * Typical implementation should first conduct verification on schema
change and parameter
+ * validity, including complex parameter interaction checks.
+ */
+ override def transformSchema(schema: StructType): StructType = {
+ transformLSHSchema(schema)
+ }
+
+ /**
+ * Given a large dataset and an item, approximately find at most k items
which have the closest
+ * distance to the item.
+ *
+ * This method has implemented two way of fetching k nearest neighbors:
+ * Single Probing: Fast, return at most k elements (Probing only one
buckets)
+ * Multiple Probing: Slow, return exact k elements (Probing multiple
buckets close to the key)
+ *
+ * @param dataset the dataset to look for the key
+ * @param key The key to hash for the item
+ * @param k The maximum number of items closest to the key
+ * @param singleProbing True for using Single Probing; false for
multiple probing
+ * @param distCol The column to store the distance between pairs
+ * @return A dataset containing at most k items closest to the key. A
distCol is added to show
+ * the distance between each record and the key.
+ */
+ def approxNearestNeighbors(dataset: Dataset[_], key: KeyType, k: Int = 1,
+ singleProbing: Boolean = true,
+ distCol: String = "distance"): Dataset[_] = {
+ assert(k > 0, "The number of nearest neighbors cannot be less than 1")
+ // Get Hash Value of the key v
+ val keyHash = hashFunction(key)
+ val modelDataset = transform(dataset)
+
+ // In the origin dataset, find the hash value u that is closest to v
+ val hashDistUDF = udf((x: Vector) => hashDistance(x, keyHash),
DataTypes.DoubleType)
+ val hashDistCol = hashDistUDF(col($(outputCol)))
+
+ val modelSubset = if (singleProbing) {
+ modelDataset.filter(hashDistCol === 0.0)
+ } else {
+ // Compute threshold to get exact k elements.
+ val modelDatasetSortedByHash =
modelDataset.sort(hashDistCol).limit(k)
+ val thresholdDataset =
modelDatasetSortedByHash.select(max(hashDistCol))
+ val hashThreshold =
thresholdDataset.collect()(0)(0).asInstanceOf[Double]
+
+ // Filter the dataset where the hash value is less than the
threshold.
+ modelDataset.filter(hashDistCol <= hashThreshold)
+ }
+
+ // Get the top k nearest neighbor by their distance to the key
+ val keyDistUDF = udf((x: KeyType) => keyDistance(x, key),
DataTypes.DoubleType)
+ val modelSubsetWithDistCol = modelSubset.withColumn(distCol,
keyDistUDF(col($(inputCol))))
+ modelSubsetWithDistCol.sort(distCol).limit(k)
+ }
+
+ /**
+ * Preprocess step for approximate similarity join. Transform and
explode the outputCol to
+ * explodeCols.
+ * @param dataset The dataset to transform and explode.
+ * @param explodeCols The alias for the exploded columns, must be a seq
of two strings.
+ * @return A dataset containing idCol, inputCol and explodeCols
+ */
+ private[this] def processDataset(dataset: Dataset[_], explodeCols:
Seq[String]): Dataset[_] = {
+ if (explodeCols.size != 2) {
+ throw new Exception("explodeCols must be two strings.")
+ }
+ val vectorToMap: UserDefinedFunction = udf((x: Vector) =>
x.asBreeze.iterator.toMap,
+ MapType(DataTypes.IntegerType, DataTypes.DoubleType))
+ transform(dataset)
+ .select(col("*"),
explode(vectorToMap(col($(outputCol)))).as(explodeCols))
+ }
+
+ /**
+ * Recreate a column using the same column name but different attribute
id. Used in approximate
+ * similarity join.
+ * @param dataset The dataset where a column need to recreate
+ * @param colName The name of the column to recreate
+ * @param tmpColName A temporary column name which does not conflict
with existing columns
+ * @return
+ */
+ private[this] def recreateCol(dataset: Dataset[_], colName: String,
+ tmpColName: String): Dataset[_] = {
+ dataset
+ .withColumnRenamed(colName, tmpColName)
+ .withColumn(colName, col(tmpColName))
+ .drop(tmpColName)
+ }
+
+ /**
+ * Join two dataset to approximately find all pairs of records whose
distance are smaller
+ * than the threshold.
+ * @param datasetA One of the datasets to join
+ * @param datasetB Another dataset to join
+ * @param threshold The threshold for the distance of record pairs
+ * @param distCol The column to store the distance between pairs
+ * @return A joined dataset containing pairs of records. A distCol is
added to show the distance
+ * between each pair of records.
+ */
+ def approxSimilarityJoin(datasetA: Dataset[_], datasetB: Dataset[_],
threshold: Double,
+ distCol: String = "distance"): Dataset[_] = {
+
+ val explodeCols = Seq("lsh#entry", "lsh#hashValue")
+ val explodedA = processDataset(datasetA, explodeCols)
+
+ // If this is a self join, we need to recreate the inputCol of
datasetB to avoid ambiguity.
+ // TODO: Remove recreateCol logic once SPARK-17154 is resolved.
+ val explodedB = if (datasetA != datasetB) {
+ processDataset(datasetB, explodeCols)
+ } else {
+ val recreatedB = recreateCol(datasetB, $(inputCol),
s"${$(inputCol)}#${Random.nextString(5)}")
+ processDataset(recreatedB, explodeCols)
+ }
+
+ // Do a hash join on where the exploded hash values are equal.
+ val joinedDataset = explodedA.join(explodedB, explodeCols)
+ .drop(explodeCols: _*)
+
+ // Add a new column to store the distance of the two records.
+ val distUDF = udf((x: KeyType, y: KeyType) => keyDistance(x, y),
DataTypes.DoubleType)
+ val joinedDatasetWithDist = joinedDataset.select(col("*"),
+ distUDF(explodedA($(inputCol)), explodedB($(inputCol))).as(distCol)
+ )
+
+ // Filter the joined datasets where the distance are smaller than the
threshold.
+ joinedDatasetWithDist.filter(col(distCol) < threshold).distinct()
+ }
+}
+
+/**
+ * Locality Sensitive Hashing for different metrics space. Support basic
transformation with a new
+ * hash column, approximate nearest neighbor search with a dataset and a
key, and approximate
+ * similarity join of two datasets.
+ *
+ * Currently the following LSH family is implemented:
+ * - Euclidean Distance: Random Projection
+ *
+ * References:
+ * (1) Gionis, Aristides, Piotr Indyk, and Rajeev Motwani. "Similarity
search in high dimensions
+ * via hashing." VLDB 7 Sep. 1999: 518-529.
+ * (2) Wang, Jingdong et al. "Hashing for similarity search: A survey."
arXiv preprint
+ * arXiv:1408.2927 (2014).
+ * @tparam KeyType The input key type of LSH
+ * @tparam T The class type of lsh
+ */
+abstract class LSH[KeyType, T <: LSHModel[KeyType, T]] extends
Estimator[T] with LSHParams {
+ /** @group setParam */
+ def setInputCol(value: String): this.type = set(inputCol, value)
--- End diff --
These methods that return `this.type` need to be overridden in the concrete
subclasses for Java compat in the method chaining style (sadly)... see
`RandomForestClassifier` for an example.
---
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
