如何将 Sparks MatrixFactorizationModel recommendProductsForUsers 保存到 Hbase

Question

我是 spark 的新手，我想将 recommendProductsForUsers 的输出保存到 Hbase table。我找到了一个示例 (https://sparkkb.wordpress.com/2015/05/04/save-javardd-to-hbase-using-saveasnewapihadoopdataset-spark-api-java-coding/)，显示使用 JavaPairRDD 和 saveAsNewAPIHadoopDataset 进行保存。

如何将 JavaRDD<Tuple2<Object, Rating[]>> 转换为 JavaPairRDD<ImmutableBytesWritable, Put> 以便我可以使用 saveAsNewAPIHadoopDataset？

//Loads the data from hdfs
    MatrixFactorizationModel sameModel = MatrixFactorizationModel.load(jsc.sc(), trainedDataPath);  

//Get recommendations for all users
    JavaRDD<Tuple2<Object, Rating[]>> ratings3 = sameModel.recommendProductsForUsers(noOfProductsToReturn).toJavaRDD();

Answer 1

通过使用 mapToPair。来自您提供示例的同一来源（我手动更改了类型）：

JavaPairRDD<ImmutableBytesWritable, Put> hbasePuts = javaRDD.mapToPair(
  new PairFunction<Tuple2<Object, Rating[]>, ImmutableBytesWritable, Put>() {
@Override
public Tuple2<ImmutableBytesWritable, Put> call(Tuple2<Object, Rating[]> row) throws Exception {

   Put put = new Put(Bytes.toBytes(row.getString(0)));
   put.add(Bytes.toBytes("columFamily"), Bytes.toBytes("columnQualifier1"), Bytes.toBytes(row.getString(1)));
   put.add(Bytes.toBytes("columFamily"), Bytes.toBytes("columnQualifier2"), Bytes.toBytes(row.getString(2)));

       return new Tuple2<ImmutableBytesWritable, Put>(new ImmutableBytesWritable(), put);     
}
 });

它是这样的，你创建一个新的 put 实例，在构造函数中为其提供行键，然后为每一列调用 add。然后你 return 创建了看跌期权。

Answer 2

这就是我解决上述问题的方法，希望这对某人有所帮助。

    JavaPairRDD<ImmutableBytesWritable, Put> hbasePuts1 = ratings3
                    .mapToPair(new PairFunction<Tuple2<Object, Rating[]>, ImmutableBytesWritable, Put>() {

                        @Override
                        public Tuple2<ImmutableBytesWritable, Put> call(Tuple2<Object, Rating[]> arg0)
                                throws Exception {
                            Rating[] userAndProducts = arg0._2;
                            System.out.println("***********" + userAndProducts.length + "**************");
                            List<Item> items = new ArrayList<Item>();
                            Put put = null
                            String recommendedProduct = "";                         
                            for (Rating r : userAndProducts) {  

//Some logic here to convert Ratings into appropriate put command
// recommendedProduct = r.product; 

}

                            put.addColumn(Bytes.toBytes("recommendation"), Bytes.toBytes("product"),Bytes.toBytes(recommendedProduct));                     Bytes.toBytes("product"),Bytes.toBytes(response.getItems().toString()));

                            return new Tuple2<ImmutableBytesWritable, Put>(new ImmutableBytesWritable(), put);
                        }
                    });

            System.out.println("*********** Number of records in JavaPairRdd: "+ hbasePuts1.count() +"**************");
            hbasePuts1.saveAsNewAPIHadoopDataset(newApiJobConfig.getConfiguration());
            jsc.stop();

Answer 3

我们刚刚开源了 Splice Machine，我们有将 MLIB 与查询和存储集成到 Splice Machine 中的示例。我不知道这是否有帮助，但我想我会让你知道。

http://community.splicemachine.com/use-spark-libraries-splice-machine/

感谢post，太棒了。

如何将 Sparks MatrixFactorizationModel recommendProductsForUsers 保存到 Hbase

How to save Sparks MatrixFactorizationModel recommendProductsForUsers to Hbase

hbase

matrix-factorization

apache-spark