比较RDD对象-Apache Spark

Question

I'm fairly new into the apache spark technology and I'm having some problems while trying to analyze data I'm pulling from my files.

I have big list of genes information, and I'm pulling that information to a RDD, so far so good.

JavaRDD<Gene> inputfile = sc.textFile(logFile).map(
        new Function<String, Gene>() {
            @Override
            public Gene call(String line) throws Exception {
                String[] values = line.split("\t");
                Gene gen = null;

                //We are only interested in genes;
                if( values.length > 2 && values[2].equalsIgnoreCase("gene") && !line.contains("#")){
                    String[] infoGene = values[8].split(";");

                    String geneId = StringUtils.substringBetween(infoGene[0], "\"");
                    String geneType = StringUtils.substringBetween(infoGene[2], "\"");
                    String geneName = StringUtils.substringBetween(infoGene[4], "\"");
                    gen = new Gene(geneName,values[3],values[4]);

                    return gen;
                }
                return gen;
            }
        }
    ).filter(new Function<Gene, Boolean>() {
        @Override
        public Boolean call(Gene gene) throws Exception {
            if(gene == null)
                return false;
            else
                return true;
        }
    });

The Gene class:

public class Gene implements Serializable{
 String firstBp;
 String lastBp;
 String name;

 public Gene(String name, String firstBp, String lastBp) {
    this.name = name;
    this.firstBp = firstBp;
    this.lastBp = lastBp;
 }

 public String getFirstBp() {
    return firstBp;
 }

 public String getLastBp() {
    return lastBp;
 }

 public String getName() {
    return name;
 }

 public String toString(){
    return name + " " + firstBp + " " + lastBp;
 }}

The problem starts here, I need to analyze if 2 Genes overlay, and for that I've made this simple utility function:

 public static Boolean isOverlay(Gene gene1, Gene gene2){
    int gene1First = Integer.parseInt(gene1.getFirstBp());
    int gene1Last = Integer.parseInt(gene1.getLastBp());
    int gene2First = Integer.parseInt(gene2.getFirstBp());
    int gene2Last = Integer.parseInt(gene2.getLastBp());

    if(gene2First >= gene1First && gene2First <= gene1Last) // FirstBp - Gene2 inside
        return true;
    else if (gene2Last >= gene1First && gene2Last <= gene1Last) // LastBP - Gene2 inside
        return true;
    else if (gene1First >= gene2First && gene1First <= gene2Last) // FirstBp - Gene1 inside
        return true;
    else if (gene1Last >= gene2First && gene1Last <= gene2Last) // LastBP - Gene1 inside
        return true;
    else
        return false;
}

Now what I'm doing and I think is wrong is transforming the RDD Object into a list by doing:

 List<Gene> genesList = inputfile.collect();

And iterate over that list to check if there are overlays and save to the file the results which is taking ages because I'm not using spark.

 List<OverlayPair> overlayPairList= new ArrayList<OverlayPair>();
 List<String> visitedGenes = new ArrayList<String>();

 for (Gene gene1 : genesList){

        for (Gene gene2 : genesList) {
            if (gene1.getName().equalsIgnoreCase(gene2.getName()) || visitedGenes.contains(gene2.getName())) {
                continue;
            }

            if (isOverlay(gene1, gene2))
                overlayPairList.add(new OverlayPair(gene1.getName(), gene2.getName()));

        }
        visitedGenes.add(gene1.getName());
    }

    JavaRDD<OverlayPair> overlayFile = sc.parallelize(overlayPairList);

    //Export the results to the file
    String outputDirectory = "/Users/joaoalmeida/Desktop/Dissertacao/sol/data/mitocondrias/feup-pp/project/data/output/overlays";
    overlayFile.coalesce(1).saveAsTextFile(outputDirectory);

The Overlay pair is basically an object with the 2 genes name.

Is there anyway to do this 2nd part while taking advantage of spark? Because the time complexity of those 2 for's its to big for the amount of data I currently have.

Answer 1

是的，您必须使用RDD.cartesian函数来获取所有对，然后才能基本上应用所编写的函数。