Examples with MatrixEntry org.apache.spark.mllib.linalg.distributed.MatrixEntry used on opensource projects

Example 1 with MatrixEntry

use of org.apache.spark.mllib.linalg.distributed.MatrixEntry 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("cp.parallel.matrixmult", "false");
        }
        MLResults results = ml.execute(script);
        if (numRegisteredOutputs >= 2) {
            String configStr = ConfigurationManager.getDMLConfig().getConfigInfo();
            if (configStr.contains("cp.parallel.matrixmult: 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;
    }
}