【Spark学习笔记】2、Java版-算子系列之filter(f:T => Boolean) : RDD[T] => RDD[T]

关注

原创最新推荐文章于 2026-09-23 15:59:02 发布 637 阅读 AI 写同款· GEO 优化 ›

Spark构建生产级用户流失预测Pipeline实战用户流失预测是典型的业务驱动型机器学习任务，其核心不在算法本身，而在于特征工程的稳定性、标签定义的业务对齐性，以及端到端pipeline的可追溯与可重现能力。基于Spark DataFrame的全链路处理，能天然保障计算口径一致性、分区裁剪效率和Delta Lake版本管理能力；而pandas UDF结合LightGBM的推理架构，则在模型表达力与工程可控性之间取得关键平衡。该方案已广泛应用于电信、SaaS与金融行业，支撑千万级用户日更特征、百维以上业务可解释特征、Top-K Lift等可行动指标交付。本文阅读详情

filter算子源码

def filter(f: T => Boolean): RDD[T] = withScope {
  val cleanF = sc.clean(f)
  new MapPartitionsRDD[T, T](
    this,
    (context, pid, iter) => iter.filter(cleanF),
    preservesPartitioning = true)
}

Java代码demo

public class FilterOperator {

	public static void main(String[] args) {
		SparkConf conf = new SparkConf().setAppName("FilterOperator").setMaster("local[2]");
		JavaSparkContext sc = new JavaSparkContext(conf);

		List<Integer> numbers = Arrays.asList(1, 2, 3, 4, 5);
		JavaRDD<Integer> numberRDD = sc.parallelize(numbers);

		/**
		 * Return a new dataset formed by selecting those elements of the source on which func returns true.
		 * 
		 * filter算子是过滤，逻辑返回true保留，false就过滤掉
		 */

		JavaRDD<Integer> results = numberRDD.filter(new Function<Integer, Boolean>() {
			private static final long serialVersionUID = 1L;

			@Override
			public Boolean call(Integer number) throws Exception {
				return number % 2 == 0;
			}
		});
		results.foreach(new VoidFunction<Integer>() {
			private static final long serialVersionUID = 1L;

			@Override
			public void call(Integer result) throws Exception {
				System.out.println(result);
			}
		});
		sc.close();
	}
}

输出