[SPARK-20687][MLLIB] mllib.Matrices.fromBreeze may crash when converting from Breeze sparse matrix

ghoto · srowen · commit 06dda1d58f86 · 2017-05-22T10:27:28.000+01:00
## What changes were proposed in this pull request? When two Breeze SparseMatrices are operated, the result matrix may contain provisional 0 values extra in rowIndices and data arrays. This causes an incoherence with the colPtrs data, but Breeze get away with this incoherence by keeping a counter of the valid data. In spark, when this matrices are converted to SparseMatrices, Sparks relies solely on rowIndices, data, and colPtrs, but these might be incorrect because of breeze internal hacks. Therefore, we need to slice both rowIndices and data, using their counter of active data This method is at least called by BlockMatrix when performing distributed block operations, causing exceptions on valid operations. See http://stackoverflow.com/questions/33528555/error-thrown-when-using-blockmatrix-add ## How was this patch tested? Added a test to MatricesSuite that verifies that the conversions are valid and that code doesn't crash. Originally the same code would crash on Spark. Bugfix for https://issues.apache.org/jira/browse/SPARK-20687 Author: Ignacio Bermudez <ignaciobermudez@gmail.com> Author: Ignacio Bermudez Corrales <icorrales@splunk.com> Closes #17940 from ghoto/bug-fix/SPARK-20687.
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala b/mllib/src/main/scala/org/apache/spark/mllib/linalg/Matrices.scala
@@ -992,7 +992,16 @@ object Matrices {
         new DenseMatrix(dm.rows, dm.cols, dm.data, dm.isTranspose)
       case sm: BSM[Double] =>
         // There is no isTranspose flag for sparse matrices in Breeze
-        new SparseMatrix(sm.rows, sm.cols, sm.colPtrs, sm.rowIndices, sm.data)
+        val nsm = if (sm.rowIndices.length > sm.activeSize) {
+          // This sparse matrix has trailing zeros.
+          // Remove them by compacting the matrix.
+          val csm = sm.copy
+          csm.compact()
+          csm
+        } else {
+          sm
+        }
+        new SparseMatrix(nsm.rows, nsm.cols, nsm.colPtrs, nsm.rowIndices, nsm.data)
       case _ =>
         throw new UnsupportedOperationException(
           s"Do not support conversion from type ${breeze.getClass.getName}.")
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/linalg/MatricesSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/linalg/MatricesSuite.scala
@@ -513,6 +513,26 @@ class MatricesSuite extends SparkFunSuite {
     Matrices.fromBreeze(sum)
   }
 
+  test("Test FromBreeze when Breeze.CSCMatrix.rowIndices has trailing zeros. - SPARK-20687") {
+    // (2, 0, 0)
+    // (2, 0, 0)
+    val mat1Brz = Matrices.sparse(2, 3, Array(0, 2, 2, 2), Array(0, 1), Array(2, 2)).asBreeze
+    // (2, 1E-15, 1E-15)
+    // (2, 1E-15, 1E-15)
+    val mat2Brz = Matrices.sparse(2, 3,
+      Array(0, 2, 4, 6),
+      Array(0, 0, 0, 1, 1, 1),
+      Array(2, 1E-15, 1E-15, 2, 1E-15, 1E-15)).asBreeze
+    val t1Brz = mat1Brz - mat2Brz
+    val t2Brz = mat2Brz - mat1Brz
+    // The following operations raise exceptions on un-patch Matrices.fromBreeze
+    val t1 = Matrices.fromBreeze(t1Brz)
+    val t2 = Matrices.fromBreeze(t2Brz)
+    // t1 == t1Brz && t2 == t2Brz
+    assert((t1.asBreeze - t1Brz).iterator.map((x) => math.abs(x._2)).sum < 1E-15)
+    assert((t2.asBreeze - t2Brz).iterator.map((x) => math.abs(x._2)).sum < 1E-15)
+  }
+
   test("row/col iterator") {
     val dm = new DenseMatrix(3, 2, Array(0, 1, 2, 3, 4, 0))
     val sm = dm.toSparse
