Spark - Word count Alternate Approach

Question

I have 1 million rows , Is there any alternate way implement word count in spark , other than traditional approach of mapping each word to 1 and then reducing it by key?

Traditional Approach :

JavaPairRDD<String, Integer> counts = textFile.flatMap(s -> 
                         Arrays.asList(SPACE.split(s)).iterator())
        .mapToPair(s -> new Tuple2<>(s, 1))
        .reduceByKey((a, b) -> a + b);

Any New Approach ?

Answer 1

There surely are many ways to do this. Here are 2:

One: flat-map and make a data frame:

JavaRDD<Row> rowRdd = spark.read()
        .textFile("loremipsum.txt")
        .javaRDD()
        .flatMap(s -> Arrays.asList(s.split(" ")).iterator())
        .map(s -> RowFactory.create(s));
spark.createDataFrame(rowRdd,
                      new StructType()
                     .add(DataTypes.createStructField("word", DataTypes.StringType, true)))
      .groupBy("word")
      .count()
      .show();

Printing something like:

+------------+-----+
|        word|count|
+------------+-----+
|         Sit|   17|
|        Elit|    6|
|   vehicula.|    2|
|       eros.|    2|
|        nam.|    3|
|   porttitor|   18|
|consectetur.|    6|
...

Bonus: Use a SQL to group (if that counts as yet a further alternative)

Two: group by word and count elements in iterables:

Map<String, Long> counts = spark.read().textFile("loremipsum.txt")
        .javaRDD()
        .flatMap(s -> Arrays.asList(s.split(" ")).iterator())
        .groupBy(i -> i)
        .aggregateByKey(0L, (id, it) -> countIterable(it), (a, b) -> a + b)

        .collect() //collection of Tuple2: you can stop here
        .stream()
        .collect(Collectors.toMap(t -> t._1, t -> t._2));

Resulting in something like:

{=50, Malesuada=4, justo.=3, potenti=2, vel.=11, purus=30, curabitur.=2...}

With countIterable being defined as:

private static <T> long countIterable(Iterable<T> it) {
    long res = 0;
    for (T t : it)
        res += 1;
    return res;
}

Which can also be implemented as

return StreamSupport.stream(it.spliterator(), false).count();