Java读取和写入Spark矢量到Hdfs

Question

I wrote Vector's (org.apache.spark.mllib.linalg.Vector) to the HDFS as the following

public void writePointsToFile(Path path, FileSystem fs, Configuration conf,
        List<Vector> points) throws IOException {

    SequenceFile.Writer writer = SequenceFile.createWriter(conf,
            Writer.file(path), Writer.keyClass(LongWritable.class),
            Writer.valueClass(Vector.class));

    long recNum = 0;

    for (Vector point : points) {
        writer.append(new LongWritable(recNum++), point);
    }
    writer.close();
}

( not sure that I used the right way to do that can't test it yet )

now I need to read this file as JavaRDD<Vector> because I want to use it in Spark Clustering K-mean but don't know how to do this.

Answer 1

Spark directly supports reading Hadoop SequenceFiles. You would do something like:

JavaSparkContext sc = new JavaSparkContext(conf);
JavaPairRDD<LongWritable, Vector> input = 
    sc.sequenceFile(fileName, LongWritable.class, Vector.class);

You then just need to convert the JavaPairRDD<LongWritable, Vector> into a JavaRDD<Vector> .

JavaRDD<Vector> out = input.map(new Function<Tuple2<LongWritable, Vector>, Vector>() {

    @Override
    public Vector call(Tuple2<LongWritable, Vector> tuple) throws Exception {
        return tuple._2();
    }
});