Search in sources :

Example 36 with MatrixMetadata

use of org.apache.sysml.api.mlcontext.MatrixMetadata in project incubator-systemml by apache.

the class GNMFTest method testGNMFWithRDMLAndJava.

@Test
public void testGNMFWithRDMLAndJava() throws IOException, DMLException, ParseException {
    System.out.println("------------ BEGIN " + TEST_NAME + " TEST {" + numRegisteredInputs + ", " + numRegisteredOutputs + "} ------------");
    this.scriptType = ScriptType.DML;
    int m = 2000;
    int n = 1500;
    int k = 50;
    int maxiter = 2;
    double Eps = Math.pow(10, -8);
    getAndLoadTestConfiguration(TEST_NAME);
    List<String> proArgs = new ArrayList<String>();
    proArgs.add(input("v"));
    proArgs.add(input("w"));
    proArgs.add(input("h"));
    proArgs.add(Integer.toString(maxiter));
    proArgs.add(output("w"));
    proArgs.add(output("h"));
    programArgs = proArgs.toArray(new String[proArgs.size()]);
    fullDMLScriptName = getScript();
    rCmd = getRCmd(inputDir(), Integer.toString(maxiter), expectedDir());
    double[][] v = getRandomMatrix(m, n, 1, 5, 0.2, System.currentTimeMillis());
    double[][] w = getRandomMatrix(m, k, 0, 1, 1, System.currentTimeMillis());
    double[][] h = getRandomMatrix(k, n, 0, 1, 1, System.currentTimeMillis());
    writeInputMatrixWithMTD("v", v, true);
    writeInputMatrixWithMTD("w", w, true);
    writeInputMatrixWithMTD("h", h, true);
    for (int i = 0; i < maxiter; i++) {
        double[][] tW = TestUtils.performTranspose(w);
        double[][] tWV = TestUtils.performMatrixMultiplication(tW, v);
        double[][] tWW = TestUtils.performMatrixMultiplication(tW, w);
        double[][] tWWH = TestUtils.performMatrixMultiplication(tWW, h);
        for (int j = 0; j < k; j++) {
            for (int l = 0; l < n; l++) {
                h[j][l] = h[j][l] * (tWV[j][l] / (tWWH[j][l] + Eps));
            }
        }
        double[][] tH = TestUtils.performTranspose(h);
        double[][] vTH = TestUtils.performMatrixMultiplication(v, tH);
        double[][] hTH = TestUtils.performMatrixMultiplication(h, tH);
        double[][] wHTH = TestUtils.performMatrixMultiplication(w, hTH);
        for (int j = 0; j < m; j++) {
            for (int l = 0; l < k; l++) {
                w[j][l] = w[j][l] * (vTH[j][l] / (wHTH[j][l] + Eps));
            }
        }
    }
    boolean oldConfig = DMLScript.USE_LOCAL_SPARK_CONFIG;
    DMLScript.USE_LOCAL_SPARK_CONFIG = true;
    RUNTIME_PLATFORM oldRT = DMLScript.rtplatform;
    try {
        DMLScript.rtplatform = RUNTIME_PLATFORM.HYBRID_SPARK;
        Script script = ScriptFactory.dmlFromFile(fullDMLScriptName);
        // set positional argument values
        for (int argNum = 1; argNum <= proArgs.size(); argNum++) {
            script.in("$" + argNum, proArgs.get(argNum - 1));
        }
        // Read two matrices through RDD and one through HDFS
        if (numRegisteredInputs >= 1) {
            JavaRDD<String> vIn = sc.sc().textFile(input("v"), 2).toJavaRDD();
            MatrixMetadata mm = new MatrixMetadata(MatrixFormat.IJV, m, n);
            script.in("V", vIn, mm);
        }
        if (numRegisteredInputs >= 2) {
            JavaRDD<String> wIn = sc.sc().textFile(input("w"), 2).toJavaRDD();
            MatrixMetadata mm = new MatrixMetadata(MatrixFormat.IJV, m, k);
            script.in("W", wIn, mm);
        }
        if (numRegisteredInputs >= 3) {
            JavaRDD<String> hIn = sc.sc().textFile(input("h"), 2).toJavaRDD();
            MatrixMetadata mm = new MatrixMetadata(MatrixFormat.IJV, k, n);
            script.in("H", hIn, mm);
        }
        // Output one matrix to HDFS and get one as RDD
        if (numRegisteredOutputs >= 1) {
            script.out("H");
        }
        if (numRegisteredOutputs >= 2) {
            script.out("W");
            ml.setConfigProperty(DMLConfig.CP_PARALLEL_OPS, "false");
        }
        MLResults results = ml.execute(script);
        if (numRegisteredOutputs >= 2) {
            String configStr = ConfigurationManager.getDMLConfig().getConfigInfo();
            if (configStr.contains("cp.parallel.ops: true"))
                Assert.fail("Configuration not updated via setConfig");
        }
        if (numRegisteredOutputs >= 1) {
            RDD<String> hOut = results.getRDDStringIJV("H");
            String fName = output("h");
            try {
                MapReduceTool.deleteFileIfExistOnHDFS(fName);
            } catch (IOException e) {
                throw new DMLRuntimeException("Error: While deleting file on HDFS");
            }
            hOut.saveAsTextFile(fName);
        }
        if (numRegisteredOutputs >= 2) {
            JavaRDD<String> javaRDDStringIJV = results.getJavaRDDStringIJV("W");
            JavaRDD<MatrixEntry> matRDD = javaRDDStringIJV.map(new StringToMatrixEntry());
            Matrix matrix = results.getMatrix("W");
            MatrixCharacteristics mcW = matrix.getMatrixMetadata().asMatrixCharacteristics();
            CoordinateMatrix coordinateMatrix = new CoordinateMatrix(matRDD.rdd(), mcW.getRows(), mcW.getCols());
            JavaPairRDD<MatrixIndexes, MatrixBlock> binaryRDD = RDDConverterUtilsExt.coordinateMatrixToBinaryBlock(sc, coordinateMatrix, mcW, true);
            JavaRDD<String> wOut = RDDConverterUtils.binaryBlockToTextCell(binaryRDD, mcW);
            String fName = output("w");
            try {
                MapReduceTool.deleteFileIfExistOnHDFS(fName);
            } catch (IOException e) {
                throw new DMLRuntimeException("Error: While deleting file on HDFS");
            }
            wOut.saveAsTextFile(fName);
        }
        runRScript(true);
        // compare matrices
        HashMap<CellIndex, Double> hmWDML = readDMLMatrixFromHDFS("w");
        HashMap<CellIndex, Double> hmHDML = readDMLMatrixFromHDFS("h");
        HashMap<CellIndex, Double> hmWR = readRMatrixFromFS("w");
        HashMap<CellIndex, Double> hmHR = readRMatrixFromFS("h");
        TestUtils.compareMatrices(hmWDML, hmWR, 0.000001, "hmWDML", "hmWR");
        TestUtils.compareMatrices(hmHDML, hmHR, 0.000001, "hmHDML", "hmHR");
    } finally {
        DMLScript.rtplatform = oldRT;
        DMLScript.USE_LOCAL_SPARK_CONFIG = oldConfig;
    }
}
Also used : MatrixBlock(org.apache.sysml.runtime.matrix.data.MatrixBlock) MLResults(org.apache.sysml.api.mlcontext.MLResults) ArrayList(java.util.ArrayList) MatrixEntry(org.apache.spark.mllib.linalg.distributed.MatrixEntry) CoordinateMatrix(org.apache.spark.mllib.linalg.distributed.CoordinateMatrix) Matrix(org.apache.sysml.api.mlcontext.Matrix) CellIndex(org.apache.sysml.runtime.matrix.data.MatrixValue.CellIndex) MatrixMetadata(org.apache.sysml.api.mlcontext.MatrixMetadata) Script(org.apache.sysml.api.mlcontext.Script) DMLScript(org.apache.sysml.api.DMLScript) MatrixIndexes(org.apache.sysml.runtime.matrix.data.MatrixIndexes) IOException(java.io.IOException) CoordinateMatrix(org.apache.spark.mllib.linalg.distributed.CoordinateMatrix) DMLRuntimeException(org.apache.sysml.runtime.DMLRuntimeException) MatrixCharacteristics(org.apache.sysml.runtime.matrix.MatrixCharacteristics) RUNTIME_PLATFORM(org.apache.sysml.api.DMLScript.RUNTIME_PLATFORM) Test(org.junit.Test)

Example 37 with MatrixMetadata

use of org.apache.sysml.api.mlcontext.MatrixMetadata in project incubator-systemml by apache.

the class MLContextTest method testDataFrameSumPYDMLDoublesWithNoIDColumn.

@Test
public void testDataFrameSumPYDMLDoublesWithNoIDColumn() {
    System.out.println("MLContextTest - DataFrame sum PYDML, doubles with no ID column");
    List<String> list = new ArrayList<String>();
    list.add("10,20,30");
    list.add("40,50,60");
    list.add("70,80,90");
    JavaRDD<String> javaRddString = sc.parallelize(list);
    JavaRDD<Row> javaRddRow = javaRddString.map(new CommaSeparatedValueStringToDoubleArrayRow());
    List<StructField> fields = new ArrayList<StructField>();
    fields.add(DataTypes.createStructField("C1", DataTypes.DoubleType, true));
    fields.add(DataTypes.createStructField("C2", DataTypes.DoubleType, true));
    fields.add(DataTypes.createStructField("C3", DataTypes.DoubleType, true));
    StructType schema = DataTypes.createStructType(fields);
    Dataset<Row> dataFrame = spark.createDataFrame(javaRddRow, schema);
    MatrixMetadata mm = new MatrixMetadata(MatrixFormat.DF_DOUBLES);
    Script script = pydml("print('sum: ' + sum(M))").in("M", dataFrame, mm);
    setExpectedStdOut("sum: 450.0");
    ml.execute(script);
}
Also used : Script(org.apache.sysml.api.mlcontext.Script) StructType(org.apache.spark.sql.types.StructType) ArrayList(java.util.ArrayList) StructField(org.apache.spark.sql.types.StructField) Row(org.apache.spark.sql.Row) MatrixMetadata(org.apache.sysml.api.mlcontext.MatrixMetadata) Test(org.junit.Test)

Example 38 with MatrixMetadata

use of org.apache.sysml.api.mlcontext.MatrixMetadata in project incubator-systemml by apache.

the class MLContextTest method testRDDGoodMetadataPYDML.

@Test
public void testRDDGoodMetadataPYDML() {
    System.out.println("MLContextTest - RDD<String> good metadata PYDML");
    List<String> list = new ArrayList<String>();
    list.add("1,1,1");
    list.add("2,2,2");
    list.add("3,3,3");
    JavaRDD<String> javaRDD = sc.parallelize(list);
    RDD<String> rdd = JavaRDD.toRDD(javaRDD);
    MatrixMetadata mm = new MatrixMetadata(3, 3, 9);
    Script script = pydml("print('sum: ' + sum(M))").in("M", rdd, mm);
    setExpectedStdOut("sum: 18.0");
    ml.execute(script);
}
Also used : Script(org.apache.sysml.api.mlcontext.Script) ArrayList(java.util.ArrayList) MatrixMetadata(org.apache.sysml.api.mlcontext.MatrixMetadata) Test(org.junit.Test)

Example 39 with MatrixMetadata

use of org.apache.sysml.api.mlcontext.MatrixMetadata in project incubator-systemml by apache.

the class MLContextTest method testJavaRDDBadMetadataPYDML.

@Test(expected = MLContextException.class)
public void testJavaRDDBadMetadataPYDML() {
    System.out.println("MLContextTest - JavaRDD<String> bad metadata PYML");
    List<String> list = new ArrayList<String>();
    list.add("1,2,3");
    list.add("4,5,6");
    list.add("7,8,9");
    JavaRDD<String> javaRDD = sc.parallelize(list);
    MatrixMetadata mm = new MatrixMetadata(1, 1, 9);
    Script script = dml("print('sum: ' + sum(M))").in("M", javaRDD, mm);
    ml.execute(script);
}
Also used : Script(org.apache.sysml.api.mlcontext.Script) ArrayList(java.util.ArrayList) MatrixMetadata(org.apache.sysml.api.mlcontext.MatrixMetadata) Test(org.junit.Test)

Example 40 with MatrixMetadata

use of org.apache.sysml.api.mlcontext.MatrixMetadata in project incubator-systemml by apache.

the class MLContextTest method testJavaRDDBadMetadataDML.

@Test(expected = MLContextException.class)
public void testJavaRDDBadMetadataDML() {
    System.out.println("MLContextTest - JavaRDD<String> bad metadata DML");
    List<String> list = new ArrayList<String>();
    list.add("1,2,3");
    list.add("4,5,6");
    list.add("7,8,9");
    JavaRDD<String> javaRDD = sc.parallelize(list);
    MatrixMetadata mm = new MatrixMetadata(1, 1, 9);
    Script script = dml("print('sum: ' + sum(M));").in("M", javaRDD, mm);
    ml.execute(script);
}
Also used : Script(org.apache.sysml.api.mlcontext.Script) ArrayList(java.util.ArrayList) MatrixMetadata(org.apache.sysml.api.mlcontext.MatrixMetadata) Test(org.junit.Test)

Aggregations

MatrixMetadata (org.apache.sysml.api.mlcontext.MatrixMetadata)72 Script (org.apache.sysml.api.mlcontext.Script)68 Test (org.junit.Test)68 ArrayList (java.util.ArrayList)60 Row (org.apache.spark.sql.Row)36 StructField (org.apache.spark.sql.types.StructField)34 StructType (org.apache.spark.sql.types.StructType)34 DenseVector (org.apache.spark.ml.linalg.DenseVector)16 Vector (org.apache.spark.ml.linalg.Vector)16 VectorUDT (org.apache.spark.ml.linalg.VectorUDT)16 MLResults (org.apache.sysml.api.mlcontext.MLResults)12 MatrixCharacteristics (org.apache.sysml.runtime.matrix.MatrixCharacteristics)10 MatrixBlock (org.apache.sysml.runtime.matrix.data.MatrixBlock)10 MatrixIndexes (org.apache.sysml.runtime.matrix.data.MatrixIndexes)10 Matrix (org.apache.sysml.api.mlcontext.Matrix)8 Tuple2 (scala.Tuple2)8 URL (java.net.URL)4 List (java.util.List)4 Tuple3 (scala.Tuple3)4 Seq (scala.collection.Seq)4