Github user feynmanliang commented on a diff in the pull request:
https://github.com/apache/spark/pull/7388#discussion_r34738272
--- Diff:
mllib/src/main/scala/org/apache/spark/ml/feature/CountVectorizerModel.scala ---
@@ -19,45 +19,135 @@ package org.apache.spark.ml.feature
import scala.collection.mutable
import org.apache.spark.annotation.Experimental
-import org.apache.spark.ml.UnaryTransformer
-import org.apache.spark.ml.param.{ParamMap, ParamValidators, IntParam}
-import org.apache.spark.ml.util.Identifiable
-import org.apache.spark.mllib.linalg.{Vectors, VectorUDT, Vector}
-import org.apache.spark.sql.types.{StringType, ArrayType, DataType}
+import org.apache.spark.ml.param._
+import org.apache.spark.ml.param.shared.{HasInputCol, HasOutputCol}
+import org.apache.spark.ml.util.{Identifiable, SchemaUtils}
+import org.apache.spark.ml.{Estimator, Model}
+import org.apache.spark.mllib.linalg.{VectorUDT, Vectors}
+import org.apache.spark.rdd.RDD
+import org.apache.spark.sql.functions._
+import org.apache.spark.sql.types._
+import org.apache.spark.sql.DataFrame
+
+/**
+ * Params for [[CountVectorizer]] and [[CountVectorizerModel]].
+ */
+private[feature] trait CountVectorizerParams extends Params with
HasInputCol with HasOutputCol {
+
+ /**
+ * size of the vocabulary.
+ * If using Estimator, CountVectorizer will build a vocabulary that only
consider the top
+ * vocabSize terms ordered by term frequency across the corpus.
+ * Default: 10000
+ * @group param
+ */
+ val vocabSize: IntParam = new IntParam(this, "vocabSize", "size of the
vocabulary")
+
+ /** @group getParam */
+ def getVocabSize: Int = $(vocabSize)
+
+ /** Validates and transforms the input schema. */
+ protected def validateAndTransformSchema(schema: StructType): StructType
= {
+ SchemaUtils.checkColumnType(schema, $(inputCol), new
ArrayType(StringType, true))
+ SchemaUtils.appendColumn(schema, $(outputCol), new VectorUDT)
+ }
+
+ override def validateParams(): Unit = {
+ require($(vocabSize) > 0, s"The vocabulary size (${$(vocabSize)}) must
be above 0.")
+ }
+}
/**
* :: Experimental ::
- * Converts a text document to a sparse vector of token counts.
- * @param vocabulary An Array over terms. Only the terms in the vocabulary
will be counted.
+ * Extracts a vocabulary from document collections and generates a
CountVectorizerModel.
*/
-@Experimental
-class CountVectorizerModel (override val uid: String, val vocabulary:
Array[String])
- extends UnaryTransformer[Seq[String], Vector, CountVectorizerModel] {
+class CountVectorizer(override val uid: String)
+ extends Estimator[CountVectorizerModel] with CountVectorizerParams {
- def this(vocabulary: Array[String]) =
- this(Identifiable.randomUID("cntVec"), vocabulary)
+ def this() = this(Identifiable.randomUID("cntVec"))
/**
- * Corpus-specific filter to ignore scarce words in a document. For each
document, terms with
- * frequency (count) less than the given threshold are ignored.
+ * The minimum number of times a token must appear in the corpus to be
included in the vocabulary
* Default: 1
* @group param
*/
- val minTermFreq: IntParam = new IntParam(this, "minTermFreq",
- "minimum frequency (count) filter used to neglect scarce words (>= 1).
For each document, " +
- "terms with frequency less than the given threshold are ignored.",
ParamValidators.gtEq(1))
+ val minCount: IntParam = new IntParam(this, "minCount",
+ "minimum number of times a token must appear in the corpus to be
included in the vocabulary."
+ , ParamValidators.gtEq(1))
+
+ /** @group getParam */
+ def getMinCount: Int = $(minCount)
/** @group setParam */
- def setMinTermFreq(value: Int): this.type = set(minTermFreq, value)
+ def setInputCol(value: String): this.type = set(inputCol, value)
- /** @group getParam */
- def getMinTermFreq: Int = $(minTermFreq)
+ /** @group setParam */
+ def setOutputCol(value: String): this.type = set(outputCol, value)
- setDefault(minTermFreq -> 1)
+ /** @group setParam */
+ def setVocabSize(value: Int): this.type = set(vocabSize, value)
+
+ /** @group setParam */
+ def setMinCount(value: Int): this.type = set(minCount, value)
- override protected def createTransformFunc: Seq[String] => Vector = {
+ setDefault(vocabSize -> 10000, minCount -> 1)
+
+ override def fit(dataset: DataFrame): CountVectorizerModel = {
+ transformSchema(dataset.schema, logging = true)
+ val input = dataset.select($(inputCol)).map(_.getAs[Seq[String]](0))
+ val min_count = $(minCount)
+ val vocab_size = $(vocabSize)
+ val wordCounts: RDD[(String, Long)] = input
+ .flatMap { case (tokens) => tokens.map(_ -> 1L) }
+ .reduceByKey(_ + _)
+ .filter(_._2 >= min_count)
+ wordCounts.cache()
+ val fullVocabSize = wordCounts.count()
+ val vocab: Array[String] = {
+ val tmpSortedWC: Array[(String, Long)] = if (fullVocabSize <=
vocab_size) {
+ // Use all terms
+ wordCounts.collect().sortBy(-_._2)
+ } else {
+ // Sort terms to select vocab
+ wordCounts.sortBy(_._2, ascending = false).take(vocab_size)
+ }
+ tmpSortedWC.map(_._1)
+ }
+
+ require(vocab.length > 0, "The vocabulary size should be > 0. Adjust
minCount as necessary.")
--- End diff --
If we do decide to throw and exception, we should add a test for it
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
