[ https://issues.apache.org/jira/browse/FLINK-3919?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15307367#comment-15307367 ]
ASF GitHub Bot commented on FLINK-3919: --------------------------------------- Github user chobeat commented on a diff in the pull request: https://github.com/apache/flink/pull/1996#discussion_r65134910 --- Diff: flink-libraries/flink-ml/src/main/scala/org/apache/flink/ml/math/distributed/DistributedRowMatrix.scala --- @@ -0,0 +1,182 @@ +/* + * 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.flink.ml.math.distributed + +import java.lang + +import breeze.linalg.{CSCMatrix => BreezeSparseMatrix, Matrix => BreezeMatrix, Vector => BreezeVector} +import org.apache.flink.api.common.functions.RichGroupReduceFunction +import org.apache.flink.api.common.typeinfo.TypeInformation +import org.apache.flink.api.scala._ +import org.apache.flink.ml.math.{Matrix => FlinkMatrix, _} +import org.apache.flink.util.Collector +import org.apache.flink.ml.math.Breeze._ +import scala.collection.JavaConversions._ + +/** + * Distributed row-major matrix representation. + * @param numRowsOpt If None, will be calculated from the DataSet. + * @param numColsOpt If None, will be calculated from the DataSet. + */ +class DistributedRowMatrix(data: DataSet[IndexedRow], + numRowsOpt: Option[Int] = None, + numColsOpt: Option[Int] = None) + extends DistributedMatrix { + + lazy val getNumRows: Int = numRowsOpt match { + case Some(rows) => rows + case None => data.count().toInt + } + + lazy val getNumCols: Int = numColsOpt match { + case Some(cols) => cols + case None => calcCols + } + + val getRowData = data + + private def calcCols: Int = + data.first(1).collect().headOption match { + case Some(vector) => vector.values.size + case None => 0 + } + + /** + * Collects the data in the form of a sequence of coordinates associated with their values. + * @return + */ + def toCOO: Seq[(Int, Int, Double)] = { + + val localRows = data.collect() + + for (IndexedRow(rowIndex, vector) <- localRows; + (columnIndex, value) <- vector) yield (rowIndex, columnIndex, value) + } + + /** + * Collects the data in the form of a SparseMatrix + * @return + */ + def toLocalSparseMatrix: SparseMatrix = { + val localMatrix = + SparseMatrix.fromCOO(this.getNumRows, this.getNumCols, this.toCOO) + require(localMatrix.numRows == this.getNumRows) + require(localMatrix.numCols == this.getNumCols) + localMatrix + } + + //TODO: convert to dense representation on the distributed matrix and collect it afterward + def toLocalDenseMatrix: DenseMatrix = this.toLocalSparseMatrix.toDenseMatrix + + /** + * Apply a high-order function to couple of rows + * @param fun + * @param other + * @return + */ + def byRowOperation(fun: (Vector, Vector) => Vector, + other: DistributedRowMatrix): DistributedRowMatrix = { + val otherData = other.getRowData + require(this.getNumCols == other.getNumCols) + require(this.getNumRows == other.getNumRows) + + + val result = this.data + .fullOuterJoin(otherData) + .where("rowIndex") + .equalTo("rowIndex")( + (left: IndexedRow, right: IndexedRow) => { + val row1 = Option(left).getOrElse(IndexedRow( + right.rowIndex, + SparseVector.fromCOO(right.values.size, List((0, 0.0))))) + val row2 = Option(right).getOrElse(IndexedRow( + left.rowIndex, + SparseVector.fromCOO(left.values.size, List((0, 0.0))))) + + IndexedRow(row1.rowIndex, fun(row1.values, row2.values)) + } + ) + new DistributedRowMatrix(result, numRowsOpt, numColsOpt) + } + + /** + * Add the matrix to another matrix. + * @param other + * @return + */ + def sum(other: DistributedRowMatrix): DistributedRowMatrix = { + val sumFunction: (Vector, Vector) => Vector = (x: Vector, y: Vector) => + (x.asBreeze + y.asBreeze).fromBreeze + this.byRowOperation(sumFunction, other) + } + + /** + * Subtracts another matrix. + * @param other + * @return + */ + def subtract(other: DistributedRowMatrix): DistributedRowMatrix = { + val subFunction: (Vector, Vector) => Vector = (x: Vector, y: Vector) => + (x.asBreeze - y.asBreeze).fromBreeze + this.byRowOperation(subFunction, other) + } +} + +object DistributedRowMatrix { + + type MatrixRowIndex = Int + + /** + * Builds a DistributedRowMatrix from a dataset in COO + * @param isSorted If false, sorts the row to properly build the matrix representation. + * If already sorted, set this parameter to true to skip sorting. + * @return + */ + def fromCOO(data: DataSet[(Int, Int, Double)], + numRows: Int, + numCols: Int, + isSorted: Boolean = false): DistributedRowMatrix = { + val vectorData: DataSet[(Int, SparseVector)] = data + .groupBy(0) + .reduceGroup(sparseRow => { + require(sparseRow.nonEmpty) + val sortedRow = + if (isSorted) { + sparseRow.toList + } else { + sparseRow.toList.sortBy(row => row._2) + } + val (indices, values) = sortedRow.map(x => (x._2, x._3)).unzip + (sortedRow.head._1, + SparseVector(numCols, indices.toArray, values.toArray)) + }) + + val zippedData = vectorData.map(x => IndexedRow(x._1.toInt, x._2)) + + new DistributedRowMatrix(zippedData, Some(numRows), Some(numCols)) + } +} + +case class IndexedRow(rowIndex: Int, values: Vector) + extends Ordered[IndexedRow] { --- End diff -- I thought it was the most elegant and generic way to sort a list of IndexedRow. If it's not the suggested way in Flink's codebase, I can remove it. > Distributed Linear Algebra: row-based matrix > -------------------------------------------- > > Key: FLINK-3919 > URL: https://issues.apache.org/jira/browse/FLINK-3919 > Project: Flink > Issue Type: New Feature > Components: Machine Learning Library > Reporter: Simone Robutti > Assignee: Simone Robutti > > Distributed matrix implementation as a DataSet of IndexedRow and related > operations -- This message was sent by Atlassian JIRA (v6.3.4#6332)