如何在Spark中创建一组ngram？

Question

我正在使用Scala从Spark 2.2数据帧列中提取Ngram，因此（在本示例中为trigram）：

val ngram = new NGram().setN(3).setInputCol("incol").setOutputCol("outcol")

如何创建一个包含1至5克的输出列？所以可能是这样的：

val ngram = new NGram().setN(1:5).setInputCol("incol").setOutputCol("outcol")

但这不起作用。我可以遍历N并为N的每个值创建新的数据帧，但这似乎效率很低。斯卡拉（Scala）挺拔的，有人能指出我正确的方向吗？

Answer 1

如果要将它们组合成向量，则可以通过zero323重写Python Answer。

import org.apache.spark.ml.feature._
import org.apache.spark.ml.Pipeline

def buildNgrams(inputCol: String = "tokens", 
                 outputCol: String = "features", n: Int = 3) = {

  val ngrams = (1 to n).map(i =>
      new NGram().setN(i)
        .setInputCol(inputCol).setOutputCol(s"${i}_grams")
  )

  val vectorizers = (1 to n).map(i =>
     new CountVectorizer()
      .setInputCol(s"${i}_grams")
      .setOutputCol(s"${i}_counts")
  )

  val assembler = new VectorAssembler()
    .setInputCols(vectorizers.map(_.getOutputCol).toArray)
    .setOutputCol(outputCol)

  new Pipeline().setStages((ngrams ++ vectorizers :+ assembler).toArray)

}

val df = Seq((1, Seq("a", "b", "c", "d"))).toDF("id", "tokens")

结果

buildNgrams().fit(df).transform(df).show(1, false)
// +---+------------+------------+---------------+--------------+-------------------------------+-------------------------+-------------------+-------------------------------------+
// |id |tokens      |1_grams     |2_grams        |3_grams       |1_counts                       |2_counts                 |3_counts           |features                             |
// +---+------------+------------+---------------+--------------+-------------------------------+-------------------------+-------------------+-------------------------------------+
// |1  |[a, b, c, d]|[a, b, c, d]|[a b, b c, c d]|[a b c, b c d]|(4,[0,1,2,3],[1.0,1.0,1.0,1.0])|(3,[0,1,2],[1.0,1.0,1.0])|(2,[0,1],[1.0,1.0])|[1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0]|
// +---+------------+------------+---------------+--------------+-------------------------------+-------------------------+-------------------+-------------------------------------+

使用UDF可能会更简单：

val ngram = udf((xs: Seq[String], n: Int) => 
  (1 to n).map(i => xs.sliding(i).filter(_.size == i).map(_.mkString(" "))).flatten)

spark.udf.register("ngram", ngram)

val ngramer =  new SQLTransformer().setStatement(
  """SELECT *, ngram(tokens, 3) AS ngrams FROM __THIS__"""
)

ngramer.transform(df).show(false)
// +---+------------+----------------------------------+
// |id |tokens      |ngrams                            |
// +---+------------+----------------------------------+
// |1  |[a, b, c, d]|[a, b, c, d, ab, bc, cd, abc, bcd]|
// +---+------------+----------------------------------+