Nebula Algorithm¶

Nebula Algorithm (Algorithm) is a Spark application based on GraphX. It uses a complete algorithm tool to perform graph computing on the data in the NebulaGraph database by submitting a Spark task. You can also programmatically use the algorithm under the lib repository to perform graph computing on DataFrame.

Version compatibility¶

The correspondence between the Nebula Algorithm release and the NebulaGraph core release is as follows.

Algorithm client	NebulaGraph
3.0-SNAPSHOT	nightly
3.0.0	3.0.0
2.6.x	2.6.x
2.5.0	2.5.0、2.5.1
2.1.0	2.0.0、2.0.1

Prerequisites¶

Before using the Nebula Algorithm, users need to confirm the following information:

The NebulaGraph services have been deployed and started. For details, see Nebula Installation.

The Spark version is 2.4.x.

The Scala version is 2.11.

(Optional) If users need to clone, compile, and package the latest Algorithm in Github, install Maven.

Limitations¶

When submitting the algorithm package directly, the data of the vertex ID must be an integer. That is, the vertex ID can be INT or String, but the data itself is an integer.

For non-integer String data, it is recommended to use the algorithm interface. You can use the dense_rank function of SparkSQL to encode the data as the Long type instead of the String type.

Graph computing outputs vertex datasets, and the algorithm results are stored in DataFrames as the properties of vertices. You can do further operations such as statistics and filtering according to your business requirements.

When writing the algorithm results into the NebulaGraph, make sure that the Tag in the corresponding graph space has properties corresponding to the algorithm result name. The corresponding properties of each algorithm are as follows.

Algorithm	Properties name	Properties type
pagerank	pagerank	double/string
louvain	louvain	int/string
kcore	kcore	int/string
labelpropagation	lpa	int/string
connectedcomponent	cc	int/string
stronglyconnectedcomponent	scc	int/string
betweenness	betweenness	double/string
shortestpath	shortestpath	string
degreestatic	degree,inDegree,outDegree	int/string
trianglecount	trianglecount	int/string
clusteringcoefficient	clustercoefficient	double/string
closeness	closeness	double/string
hanp	hanp	int/string
bfs	bfs	string
jaccard	jaccard	string
node2vec	node2vec	string

Supported algorithms¶

The graph computing algorithms supported by Nebula Algorithm are as follows.

Algorithm	Description	Scenario
PageRank	The rank of pages	Web page ranking, key node mining
Louvain	Community discovery	Community mining, hierarchical clustering
KCore	K core	Community discovery, financial risk control
LabelPropagation	Label propagation	Information spreading, advertising, and community discovery
Hanp	Label propagation advanced	Community discovery, recommendation system
ConnectedComponent	Connected component	Community discovery, island discovery
StronglyConnectedComponent	Strongly connected component	Community discovery
ShortestPath	The shortest path	Path planning, network planning
TriangleCount	Triangle counting	Network structure analysis
GraphTriangleCount	Graph triangle counting	Network structure and tightness analysis
BetweennessCentrality	Intermediate centrality	Key node mining, node influence computing
Closeness	Closeness centrality	Key node mining, node influence computing
DegreeStatic	Degree of statistical	Graph structure analysis
ClusteringCoefficient	Aggregation coefficient	Recommendation system, telecom fraud analysis
Jaccard	Jaccard similarity	Similarity computing, recommendation system
BFS	Breadth-First Search	Sequence traversal, shortest path planning
Node2Vec	-	Graph classification

Implementation methods¶

Nebula Algorithm implements the graph calculating as follows:

Read the graph data of DataFrame from the NebulaGraph database using the Nebula Spark Connector.
Transform the graph data of DataFrame to the GraphX graph.
Use graph algorithms provided by GraphX (such as PageRank) or self-implemented algorithms (such as Louvain).

For detailed implementation methods, see Scala file.

Get Nebula Algorithm¶

Compile and package¶

Clone the repository nebula-algorithm.

$ git clone -b v3.0.0 https://github.com/vesoft-inc/nebula-algorithm.git

Enter the directory nebula-algorithm.
```
$ cd nebula-algorithm
```

Compile and package.

$ mvn clean package -Dgpg.skip -Dmaven.javadoc.skip=true -Dmaven.test.skip=true

After the compilation, a similar file nebula-algorithm-3.x.x.jar is generated in the directory nebula-algorithm/target.

Download maven from the remote repository¶

Download address

How to use¶

Use algorithm interface (recommended)¶

The lib repository provides 10 common graph algorithms.

Add dependencies to the file pom.xml.

<dependency>
     <groupId>com.vesoft</groupId>
     <artifactId>nebula-algorithm</artifactId>
     <version>3.0.0</version>
</dependency>

Use the algorithm (take PageRank as an example) by filling in parameters. For more examples, see example.

Note

By default, the DataFrame that executes the algorithm sets the first column as the starting vertex, the second column as the destination vertex, and the third column as the edge weights (not the rank in the NebulaGraph).
```
val prConfig = new PRConfig(5, 1.0)
val louvainResult = PageRankAlgo.apply(spark, data, prConfig, false)
```
If your vertex IDs are Strings, see Pagerank Example for how to encoding and decoding them.

Submit the algorithm package directly¶

Note

There are limitations to use sealed packages. For example, when sinking a repository into NebulaGraph, the property name of the tag created in the sunk graph space must match the preset name in the code. The first method is recommended if the user has development skills.

Set the Configuration file.

{
  # Configurations related to Spark
  spark: {
    app: {
        name: LPA
        # The number of partitions of Spark
        partitionNum:100
    }
    master:local
  }

  data: {
    # Data source. Optional values are nebula, csv, and json.
    source: csv
    # Data sink. The algorithm result will be written into this sink. Optional values are nebula, csv, and text.
    sink: nebula
    # Whether the algorithm has a weight.
    hasWeight: false
  }

  # Configurations related to NebulaGraph
  nebula: {
    # Data source. When NebulaGraph is the data source of the graph computing, the configuration of `nebula.read` is valid.
    read: {
        # The IP addresses and ports of all Meta services. Multiple addresses are separated by commas (,). Example: "ip1:port1,ip2:port2".
        # To deploy NebulaGraph by using Docker Compose, fill in the port with which Docker Compose maps to the outside.
        # Check the status with `docker-compose ps`.
        metaAddress: "192.168.*.10:9559"
        # The name of the graph space in NebulaGraph.
        space: basketballplayer
        # Edge types in NebulaGraph. When there are multiple labels, the data of multiple edges will be merged.
        labels: ["serve"]
        # The property name of each edge type in NebulaGraph. This property will be used as the weight column of the algorithm. Make sure that it corresponds to the edge type.
        weightCols: ["start_year"]
    }

    # Data sink. When the graph computing result sinks into NebulaGraph, the configuration of `nebula.write` is valid.
    write:{
        # The IP addresses and ports of all Graph services. Multiple addresses are separated by commas (,). Example: "ip1:port1,ip2:port2".
        # To deploy by using Docker Compose, fill in the port with which Docker Compose maps to the outside.
        # Check the status with `docker-compose ps`.
        graphAddress: "192.168.*.11:9669"
        # The IP addresses and ports of all Meta services. Multiple addresses are separated by commas (,). Example: "ip1:port1,ip2:port2".
        # To deploy NebulaGraph by using Docker Compose, fill in the port with which Docker Compose maps to the outside.
        # Check the staus with `docker-compose ps`.
        metaAddress: "192.168.*.12:9559"
        user:root
        pswd:nebula
        # Before submitting the graph computing task, create the graph space and tag.
        # The name of the graph space in NebulaGraph.
        space:nb
        # The name of the tag in NebulaGraph. The graph computing result will be written into this tag. The property name of this tag is as follows.
        # PageRank: pagerank
        # Louvain: louvain
        # ConnectedComponent: cc
        # StronglyConnectedComponent: scc
        # LabelPropagation: lpa
        # ShortestPath: shortestpath
        # DegreeStatic: degree、inDegree、outDegree
        # KCore: kcore
        # TriangleCount: tranglecpunt
        # BetweennessCentrality: betweennedss
        tag:pagerank
    }
    }  

  local: {
    # Data source. When the data source is csv or json, the configuration of `local.read` is valid.
    read:{
        filePath: "hdfs://127.0.0.1:9000/edge/work_for.csv"
        # If the CSV file has a header or it is a json file, use the header. If not, use [_c0, _c1, _c2, ..., _cn] instead.
        # The header of the source VID column.
        srcId:"_c0"
        # The header of the destination VID column.
        dstId:"_c1"
        # The header of the weight column.
        weight: "_c2"
        # Whether the csv file has a header.
        header: false
        # The delimiter in the csv file.
        delimiter:","
    }

    # Data sink. When the graph computing result sinks to the csv or text file, the configuration of `local.write` is valid.
    write:{
        resultPath:/tmp/
    }
    }


  algorithm: {
    # The algorithm to execute. Optional values are as follow: 
    # pagerank, louvain, connectedcomponent, labelpropagation, shortestpaths, 
    # degreestatic, kcore, stronglyconnectedcomponent, trianglecount ,
    # betweenness, graphtriangleCount.
    executeAlgo: pagerank

    # PageRank
    pagerank: {
        maxIter: 10
        resetProb: 0.15 
    }

    # Louvain
    louvain: {
        maxIter: 20
        internalIter: 10
        tol: 0.5
    }

   # ...

}
}

Submit the graph computing task.

${SPARK_HOME}/bin/spark-submit --master <mode> --class com.vesoft.nebula.algorithm.Main <nebula-algorithm-3.0.0.jar_path> -p <application.conf_path>

Example:

${SPARK_HOME}/bin/spark-submit --master "local" --class com.vesoft.nebula.algorithm.Main /root/nebula-algorithm/target/nebula-algorithm-3.0-SNAPSHOT.jar -p /root/nebula-algorithm/src/main/resources/application.conf

Last update: March 13, 2023