Github user MLnick commented on a diff in the pull request: https://github.com/apache/spark/pull/18624#discussion_r127401088 --- Diff: mllib/src/main/scala/org/apache/spark/mllib/recommendation/MatrixFactorizationModel.scala --- @@ -286,40 +288,120 @@ object MatrixFactorizationModel extends Loader[MatrixFactorizationModel] { srcFeatures: RDD[(Int, Array[Double])], dstFeatures: RDD[(Int, Array[Double])], num: Int): RDD[(Int, Array[(Int, Double)])] = { - val srcBlocks = blockify(srcFeatures) - val dstBlocks = blockify(dstFeatures) - val ratings = srcBlocks.cartesian(dstBlocks).flatMap { case (srcIter, dstIter) => - val m = srcIter.size - val n = math.min(dstIter.size, num) - val output = new Array[(Int, (Int, Double))](m * n) + val srcBlocks = blockify(rank, srcFeatures).zipWithIndex() + val dstBlocks = blockify(rank, dstFeatures) + val ratings = srcBlocks.cartesian(dstBlocks).map { + case (((srcIds, srcFactors), index), (dstIds, dstFactors)) => + val m = srcIds.length + val n = dstIds.length + val dstIdMatrix = new Array[Int](m * num) + val scoreMatrix = Array.fill[Double](m * num)(Double.NegativeInfinity) + val pq = new BoundedPriorityQueue[(Int, Double)](num)(Ordering.by(_._2)) + + val ratings = srcFactors.transpose.multiply(dstFactors) + var i = 0 + var j = 0 + while (i < m) { + var k = 0 + while (k < n) { + pq += dstIds(k) -> ratings(i, k) + k += 1 + } + var size = pq.size + while (size > 0) { + size -= 1 + val factor = pq.poll() + dstIdMatrix(j + size) = factor._1 + scoreMatrix(j + size) = factor._2 + } + i += 1 + // pq.size maybe less than num, corner case + j += num + pq.clear() + } + (index, (srcIds, dstIdMatrix, new DenseMatrix(m, num, scoreMatrix))) + } + ratings.aggregateByKey(null: Array[Int], null: Array[Int], null: DenseMatrix)( --- End diff -- This is aggregating by `key` which in this case appears to be the "block index". What is the benefit then? Since each block will have a unique index, there would be no intermediate aggregation.
--- 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