常用图算法实现--Spark

使用Spark实现PageRank，强连通分量等图算法

PageRank

数据准备

边：

1 2
1 15
2 3
2 4
2 5
2 6
2 7
3 13
4 2
5 11
5 12
6 1
6 7
6 8
7 1
7 8
8 1
8 9
8 10
9 14
9 1
10 1
10 13
11 12
11 1
12 1
13 14
14 12
15 1

网页：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

将这两个文件放入HDFS：

hdfs dfs -mkdir input/PageRank
hdfs dfs -put links.txt input/PageRank
hdfs dfs -put pages.txt input/PageRank

编写程序

import org.apache.spark.SparkConf;
import org.apache.spark.api.java.*;
import org.apache.spark.api.java.function.Function;
import org.apache.spark.api.java.function.PairFunction;
import scala.Tuple2;

import static java.lang.Math.abs;


public class PageRank {
    private static int MaxIteration = 100;
    private static final double DAMPENING_FACTOR = 0.85;
    private static final double EPSILON = 0.0001;

    public static void main(String[] args) {
        SparkConf conf = new SparkConf().setAppName("PageRank");
        JavaSparkContext sc = new JavaSparkContext(conf);
        sc.setLogLevel("WARN");

        String linksFile = "hdfs:///user/hadoop/input/PageRank/links.txt";
        String pagesFile = "hdfs:///user/hadoop/input/PageRank/pages.txt";
        String rankFile = "hdfs:///user/hadoop/output/Graph/SparkPageRank";

        /**
         *  neighborRDD: (from, s)
         *  linksRDD: tuple (from, [to,1/m])
         *  pageRDD: vertex
         *  pageRankRDD: (point, 1/n)
         */


        JavaPairRDD neighborRDD = sc.textFile(linksFile)
                .mapToPair(
                        line -> new Tuple2<>(
                                Integer.parseInt(line.split(" ")[0]), 1))
                .reduceByKey((x, y) -> x + y);

        JavaPairRDD> linksRDD = sc.textFile(linksFile)
                .mapToPair(
                        line -> new Tuple2<>(
                                Integer.parseInt(line.split(" ")[0]),
                                Integer.parseInt(line.split(" ")[1])
                        ))
                .join(neighborRDD);

        JavaRDD pagesRDD = sc.textFile(pagesFile).map(line -> Integer.parseInt(line));
        long pageCount = pagesRDD.count();
        JavaPairRDD pageRankRDD = pagesRDD.mapToPair(
                vertex -> new Tuple2<>(vertex, 1.0 / pageCount)
        );

        int count = 0;
        while (count < MaxIteration) {
            JavaPairRDD NewPageRankRDD = linksRDD.join(pageRankRDD)
                    .mapToPair(
                            new PairFunction, Double>>, Integer, Double>() {
                                @Override
                                public Tuple2 call(Tuple2, Double>> ans) throws Exception {
//                               // [ toNode, fraction * rank]
                                    return new Tuple2<>(ans._2._1._1, ans._2._2/ans._2._1._2);
                                }
                            })
                    .reduceByKey((v1, v2) -> v1 + v2)
                    .mapValues(
                            new Function() {
                                double dampening = DAMPENING_FACTOR;
                                double randomJump = (1 - DAMPENING_FACTOR) / pageCount;

                                @Override
                                public Double call(Double value) throws Exception {
                                    value = value * dampening + randomJump;
                                    return value;
                                }
                            }
                    );
            count++;
            JavaPairRDD> compare = pageRankRDD.join(NewPageRankRDD).filter(each -> abs(each._2._1 - each._2._2) > EPSILON);
            if (compare.isEmpty() || count > MaxIteration)
                break;
            pageRankRDD = NewPageRankRDD;

        }

        pageRankRDD.saveAsTextFile(rankFile);

    }
}

思路：

1. 全部使用Lambda表达式进行，首先需要找到所有的边的条数，初始化Rank值
2. 然后使用Join进行合并，并计算下一轮Rank
3. 使用DAMPENING_FACTOR进行随机跳转

运行

spark-submit  --class PageRank PageRank-1.0.jar
hdfs dfs -cat output/Graph/SparkPageRank/*

结果为：

54622233513

ConnectedComponents

数据准备

提供基本数据集，与PageRank一样，指定顶点和边

vertices.txt

准备一些顶点，例如1-16

edges.txt

准备一些连接边：

1 2
2 3
2 4
3 5
6 7
8 9
8 10
5 11
11 12
10 13
9 14
13 14
1 15
16 1

将这两个文件放入HDFS：

hdfs dfs -mkdir input/ConnectedComponents
hdfs dfs -put edges.txt input/ConnectedComponents
hdfs dfs -put vertices.txt input/ConnectedComponents

编写程序

import org.apache.spark.SparkConf;
import org.apache.spark.api.java.JavaPairRDD;
import org.apache.spark.api.java.JavaSparkContext;
import scala.Tuple2;

import static java.lang.StrictMath.min;

public class ConnectedComponents {

    public static int MaxIteration = 100;

    public static void main(String[] args) {
        SparkConf conf = new SparkConf().setAppName("ConnectedComponents");
        JavaSparkContext sc = new JavaSparkContext(conf);
        sc.setLogLevel("WARN");

        String edgesFile = "hdfs:///user/hadoop/input/ConnectedComponents/edges.txt";
        String verticesFile = "hdfs:///user/hadoop/input/ConnectedComponents/vertices.txt";
        String outFile = "hdfs:///user/hadoop/output/Graph/SparkConnectedComponents";


        /**
         * edgesRDD: [x,y]
         * componentsRDD: [x,x] init
         */

        JavaPairRDD edgesRDD = sc.textFile(edgesFile)
                .mapToPair(
                        line -> new Tuple2<>(
                                Integer.parseInt(line.split(" ")[0]),
                                Integer.parseInt(line.split(" ")[1])
                        )
                );

        JavaPairRDD componentsRDD = sc.textFile(verticesFile)
                .mapToPair(
                        line -> new Tuple2<>(Integer.parseInt(line), Integer.parseInt(line))
                );

        int count = 0;

        while (count < MaxIteration) {
            JavaPairRDD newcomponentsRDD = componentsRDD.join(edgesRDD)
                    .mapToPair(
                            x -> new Tuple2<>(x._2._2, x._2._1)
                    )
                    .reduceByKey(
                            (v1, v2) -> min(v1, v2)
                    );

            JavaPairRDD> filterRDD = newcomponentsRDD.join(componentsRDD)
                    .filter(
                            each -> each._2._1 < each._2._2
                    );

            if (filterRDD.isEmpty())
                break;

            // update to componentsRDD
            componentsRDD = componentsRDD.leftOuterJoin(newcomponentsRDD).
                    mapValues(
                            v -> min(v._1, v._2.orElse(v._1))
                    );

            count++;
        }

        componentsRDD.saveAsTextFile(outFile);
    }
}

思路：

1. 首先需要将每个点映射成自己的强连通分支
2. 每次迭代，更新与自己相连的点的强连通分支，取最小值
3. 使用左连接更新原始的强连通分支

运行

spark-submit  --class ConnectedComponents ConnectedComponents-1.0.jar
hdfs dfs -cat output/Graph/SparkConnectedComponents/*

查看结果：

54622728559

SingleSourceShortestPaths

数据准备

首先我们需要准备边和点

边：

1 2 12.0
1 3 13.0
2 3 23.0
3 4 34.0
3 5 35.0
4 5 45.0
5 1 51.0

点：

1
2
3
4
5

将这两个文件放入HDFS：

hdfs dfs -mkdir input/SingleSourceShortestPaths
hdfs dfs -put edges.txt input/SingleSourceShortestPaths
hdfs dfs -put vertices.txt input/SingleSourceShortestPaths

编写程序

import org.apache.spark.SparkConf;
import org.apache.spark.api.java.JavaPairRDD;
import org.apache.spark.api.java.JavaSparkContext;
import scala.Tuple2;

import javax.validation.constraints.Max;

import static java.lang.StrictMath.min;

public class SingleSourceShortestPaths {
    public static int sourceVerticeID = 1;
    public static int MaxIteration = 100;

    public static void main(String[] args) throws Exception {
        SparkConf conf = new SparkConf().setAppName("ConnectedComponents");
        JavaSparkContext sc = new JavaSparkContext(conf);
        sc.setLogLevel("WARN");

        String edgesFile = "hdfs:///user/hadoop/input/SingleSourceShortestPaths/edges.txt";
        String verticesFile = "hdfs:///user/hadoop/input/SingleSourceShortestPaths/vertices.txt";
        String outFile = "hdfs:///user/hadoop/output/Graph/SparkSingleSourceShortestPaths";

        /**
         * edgesRDD: [from, to, dis ]
         * verticesRDD: [vertice, dis]
         */


        JavaPairRDD> edgesRDD = sc.textFile(edgesFile)
                .mapToPair(
                        line -> {
                            int from = Integer.parseInt(line.split(" ")[0]);
                            int to = Integer.parseInt(line.split(" ")[1]);
                            double dis = Double.parseDouble(line.split(" ")[2]);
                            return new Tuple2<>(from, new Tuple2<>(to, dis));
                        }
                );

        JavaPairRDD verticesRDD = sc.textFile(verticesFile)
                .mapToPair(
                        line -> {
                            int vertice = Integer.parseInt(line);
                            if (vertice == sourceVerticeID)
                                return new Tuple2<>(vertice, 0.0);
                            return new Tuple2<>(vertice, Double.POSITIVE_INFINITY);
                        }
                );

        int count = 0;
        while (count < MaxIteration) {
            // get new dis
            JavaPairRDD newVerticesRDD = verticesRDD
                    .join(edgesRDD)
                    .mapToPair(
                            line -> {
                                if (line._2._1 != Double.POSITIVE_INFINITY)
                                    return new Tuple2<>(line._2._2._1, line._2._1 + line._2._2._2);
                                return new Tuple2<>(line._2._2._1, Double.POSITIVE_INFINITY);
                            }
                    ).reduceByKey(
                            (v1, v2) -> min(v1, v2));

            JavaPairRDD> filterRDD = newVerticesRDD.join(verticesRDD)
                    .filter(
                            each -> each._2._1 < each._2._2);

            if (filterRDD.isEmpty())
                break;

            // update to verticesRDD
            verticesRDD = verticesRDD.leftOuterJoin(newVerticesRDD).
                    mapValues(
                            v -> min(v._1, v._2.orElse(v._1)));
        }
        verticesRDD.saveAsTextFile(outFile);

    }
}

思路：

1. 首先需要初始化每个顶点的距离，将原始点设置为0，其余设置为无穷
2. 每次迭代得到新的顶点距离，并使用reduceByKey最小化，比较是否更新
3. 然后将更新得到的顶点距离加入原始RDD中

运行

spark-submit  --class SingleSourceShortestPaths SingleSourceShortestPaths-1.0.jar
hdfs dfs -cat output/Graph/SparkSingleSourceShortestPaths/*

查看结果：

54623040420