scala - Spark 2.0 中访问向量列时出现 MatchError

Question

我正在尝试在 JSON 文件上创建 LDA 模型。

使用 JSON 文件创建 Spark 上下文:

import org.apache.spark.sql.SparkSession

val sparkSession = SparkSession.builder
  .master("local")
  .appName("my-spark-app")
  .config("spark.some.config.option", "config-value")
  .getOrCreate()

 val df = spark.read.json("dbfs:/mnt/JSON6/JSON/sampleDoc.txt")

显示df应该显示DataFrame

display(df)

对文本进行标记

import org.apache.spark.ml.feature.RegexTokenizer

// Set params for RegexTokenizer
val tokenizer = new RegexTokenizer()
                .setPattern("[\\W_]+")
                .setMinTokenLength(4) // Filter away tokens with length < 4
                .setInputCol("text")
                .setOutputCol("tokens")

// Tokenize document
val tokenized_df = tokenizer.transform(df)

这应该显示tokenized_df

display(tokenized_df)

获取停用词

%sh wget http://ir.dcs.gla.ac.uk/resources/linguistic_utils/stop_words > -O /tmp/stopwords

可选:将停用词复制到 tmp 文件夹

%fs cp file:/tmp/stopwords dbfs:/tmp/stopwords

收集所有停用词

val stopwords = sc.textFile("/tmp/stopwords").collect()

过滤掉停用词

 import org.apache.spark.ml.feature.StopWordsRemover

 // Set params for StopWordsRemover
 val remover = new StopWordsRemover()
                   .setStopWords(stopwords) // This parameter is optional
                   .setInputCol("tokens")
                   .setOutputCol("filtered")

 // Create new DF with Stopwords removed
 val filtered_df = remover.transform(tokenized_df)

显示过滤后的df应该验证停用词已被删除

 display(filtered_df)

向量化单词出现的频率

 import org.apache.spark.mllib.linalg.Vectors
 import org.apache.spark.sql.Row
 import org.apache.spark.ml.feature.CountVectorizer

 // Set params for CountVectorizer
 val vectorizer = new CountVectorizer()
               .setInputCol("filtered")
               .setOutputCol("features")
               .fit(filtered_df)

验证矢量化器

 vectorizer.transform(filtered_df)
           .select("id", "text","features","filtered").show()

在此之后，我发现在 LDA 中安装此向量化器时出现问题。我认为 CountVectorizer 的问题是给出稀疏向量，但 LDA 需要密集向量。仍在尝试找出问题所在。

这是 map 无法转换的异常(exception)情况。

import org.apache.spark.mllib.linalg.Vector
val ldaDF = countVectors.map { 
             case Row(id: String, countVector: Vector) => (id, countVector) 
            }
display(ldaDF)

异常:

org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 in stage 4083.0 failed 4 times, most recent failure: Lost task 0.3 in stage 4083.0 (TID 15331, 10.209.240.17): scala.MatchError: [0,(1252,[13,17,18,20,30,37,45,50,51,53,63,64,96,101,108,125,174,189,214,221,224,227,238,268,291,309,328,357,362,437,441,455,492,493,511,528,561,613,619,674,764,823,839,980,1098,1143],[1.0,1.0,2.0,1.0,1.0,1.0,2.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,3.0,1.0,2.0,1.0,5.0,1.0,2.0,2.0,1.0,4.0,1.0,2.0,3.0,1.0,1.0,1.0,1.0,1.0,2.0,1.0,1.0,1.0,1.0,1.0,2.0,1.0,2.0,1.0,1.0,1.0])] (of class org.apache.spark.sql.catalyst.expressions.GenericRowWithSchema)

LDA 有一个工作示例，不会引发任何问题

import org.apache.spark.mllib.linalg.Vectors
import org.apache.spark.sql.Row
import org.apache.spark.mllib.linalg.Vector
import org.apache.spark.mllib.clustering.{DistributedLDAModel, LDA}

val a = Vectors.dense(Array(1.0,2.0,3.0))
val b = Vectors.dense(Array(3.0,4.0,5.0))
val df = Seq((1L,a),(2L,b),(2L,a)).toDF

val ldaDF = df.map { case Row(id: Long, countVector: Vector) => (id, countVector) } 

val model = new LDA().setK(3).run(ldaDF.javaRDD)
display(df)

唯一的区别是在第二个片段中我们有一个密集矩阵。

Answer 1

这与稀疏性无关。自 Spark 2.0.0 ML Transformers 不再生成 o.a.s.mllib.linalg.VectorUDT 而是生成 o.a.s.ml.linalg.VectorUDT 并在本地映射到子类o.a.s.ml.linalg.Vector。这些与旧的 MLLib API 不兼容，旧的 MLLib API 在 Spark 2.0.0 中即将弃用。

您可以使用 Vectors.fromML 在“旧”之间进行转换:

import org.apache.spark.mllib.linalg.{Vectors => OldVectors}
import org.apache.spark.ml.linalg.{Vectors => NewVectors}

OldVectors.fromML(NewVectors.dense(1.0, 2.0, 3.0))
OldVectors.fromML(NewVectors.sparse(5, Seq(0 -> 1.0, 2 -> 2.0, 4 -> 3.0)))

但如果您已经使用 ML 转换器，那么使用 LDA 的 ML 实现更有意义。

为了方便起见，您可以使用隐式转换:

import scala.languageFeature.implicitConversions

object VectorConversions {
  import org.apache.spark.mllib.{linalg => mllib}
  import org.apache.spark.ml.{linalg => ml}

  implicit def toNewVector(v: mllib.Vector) = v.asML
  implicit def toOldVector(v: ml.Vector) = mllib.Vectors.fromML(v)
}