Better API for Matrix Market files

[fommil]

Original author: Sam.Hall...@gmail.com (May 16, 2009 11:22:13)

The API for reading and writing Matrix Market files is both cumbersome and confusing. A simple API
would be a welcome improvement.

Original issue: http://code.google.com/p/matrix-toolkits-java/issues/detail?id=16

Want to back this issue? Post a bounty on it! We accept bounties via Bountysource.

[fommil]

From rainer.k...@gmail.com on January 17, 2013 11:20:32
maybe something along these lines ? this was a quick hack so i do to not lose too much precision...
...and it does not conform to the matrix-market format, as it lacks the MatrixInfo and MatrixSize stuff (i did not need it after all, as i had just the precision issue)

fortunately java's Double.toString() and its complement Double.parseDouble() do all the work :)

hth, rainer

// own implementation, as the matrix market stuff from mtj loses precision
private static void saveMatrixSPARSE(File outputFile, Matrix M) throws FileNotFoundException {
PrintWriter pwriter = new PrintWriter(new BufferedOutputStream(new FileOutputStream(outputFile)));

    Matrix SM = (FlexCompRowMatrix) M;

    pwriter.println(String.format("%d, %d", SM.numRows(), SM.numColumns()));

    for (MatrixEntry e : SM) {
        pwriter.println(String.format("%d, %d, %s", e.row(), e.column(), Double.toString(e.get())));
    }

    pwriter.flush();
    pwriter.close();
}

// same...
private static Matrix loadMatrixFromSPARSE(File inputFile) throws IOException {
    BufferedReader reader = new BufferedReader(new InputStreamReader(new FileInputStream(inputFile), MatrixOps.UTF8_ENCODING));

    String line = reader.readLine();
    StringTokenizer tokens = new StringTokenizer(line, MatrixOps.CSV_DELIMITERS, false);
    int numRows = Integer.parseInt(tokens.nextToken());
    int numColumns = Integer.parseInt(tokens.nextToken());

    FlexCompRowMatrix M = new FlexCompRowMatrix(numRows, numColumns);

    while ((line = reader.readLine()) != null) {
        tokens = new StringTokenizer(line, MatrixOps.CSV_DELIMITERS, false);

        String rawRow = tokens.nextToken();
        String rawCol = tokens.nextToken();
        String rawD = tokens.nextToken();

        M.set(Integer.parseInt(rawRow), Integer.parseInt(rawCol), Double.parseDouble(rawD));
    }


    return M;
}

[fommil]

From Sam.Hall...@gmail.com on January 17, 2013 12:09:50
Ha! Cool. It will take me some time to get around to this, but it's really excellent that you've posted some code here as it means people can C&P as a workaround.