java - 如何使用 RowMatrix.columnSimilarities 的输出

Question

我需要计算一行各列之间的相似度，并尝试使用 columnsimilarities() 方法来获得结果。

public static void main(String[] args) {

    SparkConf sparkConf = new SparkConf().setAppName("CollarberativeFilter").setMaster("local");
        JavaSparkContext sc = new JavaSparkContext(sparkConf);
        SparkSession spark = SparkSession.builder().appName("CollarberativeFilter").getOrCreate();
        double[][] array = {{5,0,5}, {0,10,0}, {5,0,5}};
        LinkedList<Vector> rowsList = new LinkedList<Vector>();
        for (int i = 0; i < array.length; i++) {
          Vector currentRow = Vectors.dense(array[i]);
          rowsList.add(currentRow);
        }
        JavaRDD<Vector> rows = sc.parallelize(rowsList);

        // Create a RowMatrix from JavaRDD<Vector>.
        RowMatrix mat = new RowMatrix(rows.rdd());
         CoordinateMatrix simsPerfect = mat.columnSimilarities();
         RowMatrix mat2 = simsPerfect.toRowMatrix();
         List<Vector> vs2 = mat2.rows().toJavaRDD().collect();
         List<Vector> vs = mat.rows().toJavaRDD().collect();
         System.out.println("mat");
         for(Vector v: vs) {
             System.out.println(v);
         }
         System.out.println("mat2");
         for(Vector v: vs2) {
             System.out.println(v);
         }
         JavaRDD<MatrixEntry> entries = simsPerfect.entries().toJavaRDD();
         JavaRDD<String> output = entries.map(new Function<MatrixEntry, String>() {
             public String call(MatrixEntry e) {
                 return String.format("%d,%d,%s", e.i(), e.j(), e.value());
             }
         });
         output.saveAsTextFile("resources123/data.txt");

}

但是

output in the text file was 0,2,0.9999999999999998

.

接下来，我使用 double[][] array = {{1,3}, {2,7}}; 尝试了相同的示例 然后是

output of the text file was 0,1,0.9982743731749959

有人能给我解释一下答案格式吗？我不能为矩阵的每一对列都打分吗？比如在 3 x 3 矩阵中，我需要 3 个分数来表示 1,2 列、2,3 之间的相似性列，3,1 列。 任何帮助表示赞赏。

Answer 1

列相似度是用定义如下的 Cosine Similarity 计算的:

因为你包含了 scala 标签，我将作弊并重复你在 Scala REPL 中所做的事情:

scala> import org.apache.spark.mllib.linalg.{Vectors, Vector}
import org.apache.spark.mllib.linalg.{Vectors, Vector}

scala> import org.apache.spark.mllib.linalg.distributed.RowMatrix
import org.apache.spark.mllib.linalg.distributed.RowMatrix

scala> val matVec = Vector(Vectors.dense(5,0,5), Vectors.dense(0,10,0), Vectors.dense(5,0,5))
matVec: scala.collection.immutable.Vector[org.apache.spark.mllib.linalg.Vector] = Vector([5.0,0.0,5.0], [0.0,10.0,0.0], [5.0,0.0,5.0])

scala> val matRDD = sc.parallelize(matVec)
matRDD: org.apache.spark.rdd.RDD[org.apache.spark.mllib.linalg.Vector] = ParallelCollectionRDD[44] at parallelize at <console>:37

scala> val myRowMat = new RowMatrix(matRDD)
myRowMat: org.apache.spark.mllib.linalg.distributed.RowMatrix = org.apache.spark.mllib.linalg.distributed.RowMatrix@7a7a07c2

scala> myRowMat.columnSimilarities.entries.collect.foreach{println}
MatrixEntry(0,2,0.9999999999999998)

此输出意味着 (row0, col2) 只有一个非零条目。因此实际的(上三角)输出是:

0    0    .9999
0    0    0
0    0    0

这是您所期望的(因为 col0 和 col1 之间的点积为零，而 col1 和 之间的点积>col2 为零)

这是一个列相似度矩阵不太稀疏的示例:

scala> def randVec(len: Int) : org.apache.spark.mllib.linalg.Vector =
     | Vectors.dense(Array.fill(len)(Random.nextDouble))
randVec: (len: Int)org.apache.spark.mllib.linalg.Vector

scala> val randRDD = sc.parallelize(Seq.fill(3)(randVec(4))
randRDD: org.apache.spark.rdd.RDD[org.apache.spark.mllib.linalg.Vector] = ParallelCollectionRDD[123] at parallelize at <console>:38

scala> val randRowMat = new RowMatrix(randRDD)
randRowMat: org.apache.spark.mllib.linalg.distributed.RowMatrix = org.apache.spark.mllib.linalg.distributed.RowMatrix@77d9112e

scala> randRowMat.rows.collect.foreach{println}
[0.11049508671100228,0.6560383649078886,0.08647831963379027,0.918734774579884]
[0.5709766390994561,0.5404121150599919,0.8206115742925799,0.12848224469499103]
[0.5414651842028494,0.26273347471310016,0.3139446375461201,0.351113866208812]

scala> randRowMat.columnSimilarities.entries.collect.foreach{println}
MatrixEntry(0,3,0.4630854334046888)
MatrixEntry(0,2,0.9238294198864545)
MatrixEntry(2,3,0.33700154742702093)
MatrixEntry(0,1,0.7402725425024911)
MatrixEntry(1,2,0.7418690274112878)
MatrixEntry(1,3,0.8662504236158493)

代表以下矩阵:

0       0.74027     0.92382     0.46308
0       0           0.74186     0.86625
0       0           0           0.33700
0       0           0           0