Java 8 version can look like that:

        try (JavaSparkContext sc = new JavaSparkContext(configLocalMode("NaiveBayes exploiting TFIDF for spam classification in Java 8"))) {
            HashingTF hash = new HashingTF(100000);
            JavaRDD<String> file = localFile("sms-labeled.txt", sc);            
            JavaRDD<Tuple3<String, List<String>, Vector>> raw = file.distinct().map(
                s -> s.split("\\t+")
            ).map(
                a -> new Tuple2<>(a[0], Arrays.stream(a[1].split("\\s+")).map(w -> w.toLowerCase()).collect(Collectors.toList()))
            ).map(
                t -> new Tuple3<>(t._1, t._2, hash.transform(t._2))
            ).cache();

            IDFModel idf = new IDF().fit(raw.map(t -> t._3()).rdd());
            JavaRDD<LabeledPoint> data = raw.map(t -> {
                int label = 0;
                if(t._1().equals("spam")){
                    label = 1;
                }
                return new LabeledPoint(label, idf.transform(t._3()));
            });

            JavaRDD<LabeledPoint>[] split = data.randomSplit(new double[] { .8, .2 });
            JavaRDD<LabeledPoint> train = split[0].cache();
            JavaRDD<LabeledPoint> test = split[1].cache();

            NaiveBayesModel model = NaiveBayes.train(train.rdd());
            evaluateModel(model, test);            
        }

And method for model evaluation:

        test.take(5).forEach(x -> {
            System.out.println(String.format("Predicted: %.1f, Label: %.1f", model.predict(x.features()), x.label()));    
        });

        JavaPairRDD<Object, Object> predictionsAndLabels = test.mapToPair(
            p -> new Tuple2<Object, Object>(model.predict(p.features()), p.label())
        );

        Stats stats = Stats.apply(confusionMatrix(predictionsAndLabels.rdd()));
        System.out.println(stats.toString());

        BinaryClassificationMetrics metrics = new BinaryClassificationMetrics(predictionsAndLabels.rdd());
        printMetrics(metrics);