First attempt to resolve issue with inferring func types in 2.12 by instead using info captured when UDF is registered -- capturing which types are nullable (i.e. not primitive)

srowen · srowen · commit 721860d38b88 · 2018-08-24T17:08:13.000-05:00
diff --git a/core/src/main/scala/org/apache/spark/api/python/PythonRunner.scala b/core/src/main/scala/org/apache/spark/api/python/PythonRunner.scala
@@ -239,7 +239,7 @@ private[spark] abstract class BasePythonRunner[IN, OUT](
         }
         // Close ServerSocket on task completion.
         serverSocket.foreach { server =>
-          context.addTaskCompletionListener(_ => server.close())
+          context.addTaskCompletionListener[Unit](_ => server.close())
         }
         val boundPort: Int = serverSocket.map(_.getLocalPort).getOrElse(0)
         if (boundPort == -1) {
diff --git a/mllib/src/main/scala/org/apache/spark/ml/Predictor.scala b/mllib/src/main/scala/org/apache/spark/ml/Predictor.scala
@@ -211,8 +211,8 @@ abstract class PredictionModel[FeaturesType, M <: PredictionModel[FeaturesType,
   }
 
   protected def transformImpl(dataset: Dataset[_]): DataFrame = {
-    val predictUDF = udf { (features: Any) =>
-      predict(features.asInstanceOf[FeaturesType])
+    val predictUDF = udfInternal { features: FeaturesType =>
+      predict(features)
     }
     dataset.withColumn($(predictionCol), predictUDF(col($(featuresCol))))
   }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/Classifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/Classifier.scala
@@ -18,7 +18,7 @@
 package org.apache.spark.ml.classification
 
 import org.apache.spark.SparkException
-import org.apache.spark.annotation.{DeveloperApi, Since}
+import org.apache.spark.annotation.DeveloperApi
 import org.apache.spark.ml.{PredictionModel, Predictor, PredictorParams}
 import org.apache.spark.ml.feature.LabeledPoint
 import org.apache.spark.ml.linalg.{Vector, VectorUDT}
@@ -164,8 +164,8 @@ abstract class ClassificationModel[FeaturesType, M <: ClassificationModel[Featur
     var outputData = dataset
     var numColsOutput = 0
     if (getRawPredictionCol != "") {
-      val predictRawUDF = udf { (features: Any) =>
-        predictRaw(features.asInstanceOf[FeaturesType])
+      val predictRawUDF = udfInternal { features: FeaturesType =>
+        predictRaw(features)
       }
       outputData = outputData.withColumn(getRawPredictionCol, predictRawUDF(col(getFeaturesCol)))
       numColsOutput += 1
@@ -174,8 +174,8 @@ abstract class ClassificationModel[FeaturesType, M <: ClassificationModel[Featur
       val predUDF = if (getRawPredictionCol != "") {
         udf(raw2prediction _).apply(col(getRawPredictionCol))
       } else {
-        val predictUDF = udf { (features: Any) =>
-          predict(features.asInstanceOf[FeaturesType])
+        val predictUDF = udfInternal { features: FeaturesType =>
+          predict(features)
         }
         predictUDF(col(getFeaturesCol))
       }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala
@@ -287,8 +287,8 @@ class GBTClassificationModel private[ml](
 
   override protected def transformImpl(dataset: Dataset[_]): DataFrame = {
     val bcastModel = dataset.sparkSession.sparkContext.broadcast(this)
-    val predictUDF = udf { (features: Any) =>
-      bcastModel.value.predict(features.asInstanceOf[Vector])
+    val predictUDF = udfInternal { features: Vector =>
+      bcastModel.value.predict(features)
     }
     dataset.withColumn($(predictionCol), predictUDF(col($(featuresCol))))
   }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/ProbabilisticClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/ProbabilisticClassifier.scala
@@ -113,8 +113,8 @@ abstract class ProbabilisticClassificationModel[
     var outputData = dataset
     var numColsOutput = 0
     if ($(rawPredictionCol).nonEmpty) {
-      val predictRawUDF = udf { (features: Any) =>
-        predictRaw(features.asInstanceOf[FeaturesType])
+      val predictRawUDF = udfInternal { features: FeaturesType =>
+        predictRaw(features)
       }
       outputData = outputData.withColumn(getRawPredictionCol, predictRawUDF(col(getFeaturesCol)))
       numColsOutput += 1
@@ -123,8 +123,8 @@ abstract class ProbabilisticClassificationModel[
       val probUDF = if ($(rawPredictionCol).nonEmpty) {
         udf(raw2probability _).apply(col($(rawPredictionCol)))
       } else {
-        val probabilityUDF = udf { (features: Any) =>
-          predictProbability(features.asInstanceOf[FeaturesType])
+        val probabilityUDF = udfInternal { features: FeaturesType =>
+          predictProbability(features)
         }
         probabilityUDF(col($(featuresCol)))
       }
@@ -137,8 +137,8 @@ abstract class ProbabilisticClassificationModel[
       } else if ($(probabilityCol).nonEmpty) {
         udf(probability2prediction _).apply(col($(probabilityCol)))
       } else {
-        val predictUDF = udf { (features: Any) =>
-          predict(features.asInstanceOf[FeaturesType])
+        val predictUDF = udfInternal { features: FeaturesType =>
+          predict(features)
         }
         predictUDF(col($(featuresCol)))
       }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala b/mllib/src/main/scala/org/apache/spark/ml/classification/RandomForestClassifier.scala
@@ -209,8 +209,8 @@ class RandomForestClassificationModel private[ml] (
 
   override protected def transformImpl(dataset: Dataset[_]): DataFrame = {
     val bcastModel = dataset.sparkSession.sparkContext.broadcast(this)
-    val predictUDF = udf { (features: Any) =>
-      bcastModel.value.predict(features.asInstanceOf[Vector])
+    val predictUDF = udfInternal { features: Vector =>
+      bcastModel.value.predict(features)
     }
     dataset.withColumn($(predictionCol), predictUDF(col($(featuresCol))))
   }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/FeatureHasher.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/FeatureHasher.scala
@@ -165,7 +165,7 @@ class FeatureHasher(@Since("2.3.0") override val uid: String) extends Transforme
       }
     }
 
-    val hashFeatures = udf { row: Row =>
+    val hashFeatures = udfInternal { row: Row =>
       val map = new OpenHashMap[Int, Double]()
       localInputCols.foreach { colName =>
         val fieldIndex = row.fieldIndex(colName)
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/HashingTF.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/HashingTF.scala
@@ -25,7 +25,7 @@ import org.apache.spark.ml.param.shared.{HasInputCol, HasOutputCol}
 import org.apache.spark.ml.util._
 import org.apache.spark.mllib.feature
 import org.apache.spark.sql.{DataFrame, Dataset}
-import org.apache.spark.sql.functions.{col, udf}
+import org.apache.spark.sql.functions.{col, udfInternal}
 import org.apache.spark.sql.types.{ArrayType, StructType}
 
 /**
@@ -95,7 +95,7 @@ class HashingTF @Since("1.4.0") (@Since("1.4.0") override val uid: String)
     val outputSchema = transformSchema(dataset.schema)
     val hashingTF = new feature.HashingTF($(numFeatures)).setBinary($(binary))
     // TODO: Make the hashingTF.transform natively in ml framework to avoid extra conversion.
-    val t = udf { terms: Seq[_] => hashingTF.transform(terms).asML }
+    val t = udfInternal { terms: Seq[_] => hashingTF.transform(terms).asML }
     val metadata = outputSchema($(outputCol)).metadata
     dataset.select(col("*"), t(col($(inputCol))).as($(outputCol), metadata))
   }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/Interaction.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/Interaction.scala
@@ -73,7 +73,7 @@ class Interaction @Since("1.6.0") (@Since("1.6.0") override val uid: String) ext
     val featureEncoders = getFeatureEncoders(inputFeatures)
     val featureAttrs = getFeatureAttrs(inputFeatures)
 
-    def interactFunc = udf { row: Row =>
+    def interactFunc = udfInternal { row: Row =>
       var indices = ArrayBuilder.make[Int]
       var values = ArrayBuilder.make[Double]
       var size = 1
diff --git a/mllib/src/main/scala/org/apache/spark/ml/feature/VectorAssembler.scala b/mllib/src/main/scala/org/apache/spark/ml/feature/VectorAssembler.scala
@@ -140,7 +140,7 @@ class VectorAssembler @Since("1.4.0") (@Since("1.4.0") override val uid: String)
       case VectorAssembler.ERROR_INVALID => (dataset, false)
     }
     // Data transformation.
-    val assembleFunc = udf { r: Row =>
+    val assembleFunc = udfInternal { r: Row =>
       VectorAssembler.assemble(lengths, keepInvalid)(r.toSeq: _*)
     }.asNondeterministic()
     val args = $(inputCols).map { c =>
diff --git a/mllib/src/main/scala/org/apache/spark/ml/recommendation/ALS.scala b/mllib/src/main/scala/org/apache/spark/ml/recommendation/ALS.scala
@@ -85,7 +85,7 @@ private[recommendation] trait ALSModelParams extends Params with HasPredictionCo
    * Attempts to safely cast a user/item id to an Int. Throws an exception if the value is
    * out of integer range or contains a fractional part.
    */
-  protected[recommendation] val checkedCast = udf { (n: Any) =>
+  protected[recommendation] val checkedCast = udfInternal { n: Any =>
     n match {
       case v: Int => v // Avoid unnecessary casting
       case v: Number =>
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/GBTRegressor.scala
@@ -34,7 +34,6 @@ import org.apache.spark.ml.util.DefaultParamsReader.Metadata
 import org.apache.spark.ml.util.Instrumentation.instrumented
 import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo}
 import org.apache.spark.mllib.tree.model.{GradientBoostedTreesModel => OldGBTModel}
-import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.{DataFrame, Dataset, Row}
 import org.apache.spark.sql.functions._
 
@@ -245,8 +244,8 @@ class GBTRegressionModel private[ml](
 
   override protected def transformImpl(dataset: Dataset[_]): DataFrame = {
     val bcastModel = dataset.sparkSession.sparkContext.broadcast(this)
-    val predictUDF = udf { (features: Any) =>
-      bcastModel.value.predict(features.asInstanceOf[Vector])
+    val predictUDF = udfInternal { features: Vector =>
+      bcastModel.value.predict(features)
     }
     dataset.withColumn($(predictionCol), predictUDF(col($(featuresCol))))
   }
diff --git a/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala b/mllib/src/main/scala/org/apache/spark/ml/regression/RandomForestRegressor.scala
@@ -195,8 +195,8 @@ class RandomForestRegressionModel private[ml] (
 
   override protected def transformImpl(dataset: Dataset[_]): DataFrame = {
     val bcastModel = dataset.sparkSession.sparkContext.broadcast(this)
-    val predictUDF = udf { (features: Any) =>
-      bcastModel.value.predict(features.asInstanceOf[Vector])
+    val predictUDF = udfInternal { features: Vector =>
+      bcastModel.value.predict(features)
     }
     dataset.withColumn($(predictionCol), predictUDF(col($(featuresCol))))
   }
diff --git a/project/MimaExcludes.scala b/project/MimaExcludes.scala
@@ -36,6 +36,11 @@ object MimaExcludes {
 
   // Exclude rules for 2.4.x
   lazy val v24excludes = v23excludes ++ Seq(
+    // [SPARK-25044] Address translation of LMF closure primitive args to Object in Scala 2.12
+    ProblemFilters.exclude[MissingTypesProblem]("org.apache.spark.sql.expressions.UserDefinedFunction$"),
+    ProblemFilters.exclude[DirectMissingMethodProblem]("org.apache.spark.sql.expressions.UserDefinedFunction.apply"),
+    ProblemFilters.exclude[DirectMissingMethodProblem]("org.apache.spark.sql.expressions.UserDefinedFunction.copy"),
+
     // [SPARK-24296][CORE] Replicate large blocks as a stream.
     ProblemFilters.exclude[DirectMissingMethodProblem]("org.apache.spark.network.netty.NettyBlockRpcServer.this"),
     // [SPARK-23528] Add numIter to ClusteringSummary
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/ScalaReflection.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/ScalaReflection.scala
@@ -932,15 +932,6 @@ trait ScalaReflection {
     tpe.dealias.erasure.typeSymbol.asClass.fullName
   }
 
-  /**
-   * Returns classes of input parameters of scala function object.
-   */
-  def getParameterTypes(func: AnyRef): Seq[Class[_]] = {
-    val methods = func.getClass.getMethods.filter(m => m.getName == "apply" && !m.isBridge)
-    assert(methods.length == 1)
-    methods.head.getParameterTypes
-  }
-
   /**
    * Returns the parameter names and types for the primary constructor of this type.
    *
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
@@ -2149,28 +2149,36 @@ class Analyzer(
 
       case p => p transformExpressionsUp {
 
-        case udf @ ScalaUDF(func, _, inputs, _, _, _, _) =>
-          val parameterTypes = ScalaReflection.getParameterTypes(func)
-          assert(parameterTypes.length == inputs.length)
-
-          // TODO: skip null handling for not-nullable primitive inputs after we can completely
-          // trust the `nullable` information.
-          // (cls, expr) => cls.isPrimitive && expr.nullable
-          val needsNullCheck = (cls: Class[_], expr: Expression) =>
-            cls.isPrimitive && !expr.isInstanceOf[KnownNotNull]
-          val inputsNullCheck = parameterTypes.zip(inputs)
-            .filter { case (cls, expr) => needsNullCheck(cls, expr) }
-            .map { case (_, expr) => IsNull(expr) }
-            .reduceLeftOption[Expression]((e1, e2) => Or(e1, e2))
-          // Once we add an `If` check above the udf, it is safe to mark those checked inputs
-          // as not nullable (i.e., wrap them with `KnownNotNull`), because the null-returning
-          // branch of `If` will be called if any of these checked inputs is null. Thus we can
-          // prevent this rule from being applied repeatedly.
-          val newInputs = parameterTypes.zip(inputs).map{ case (cls, expr) =>
-            if (needsNullCheck(cls, expr)) KnownNotNull(expr) else expr }
-          inputsNullCheck
-            .map(If(_, Literal.create(null, udf.dataType), udf.copy(children = newInputs)))
-            .getOrElse(udf)
+        case udf@ScalaUDF(func, _, inputs, _, _, _, _, nullableTypes) =>
+
+          if (nullableTypes.isEmpty) {
+            // If no nullability info is available, do nothing. No fields will be specially
+            // checked for null in the plan. If nullability info is incorrect, the results
+            // of the UDF could be wrong.
+            udf
+
+          } else {
+            // Otherwise, add special handling of null for fields that can't accept null.
+            // The result of operations like this, when passed null, is generally to return null.
+            assert(nullableTypes.length == inputs.length)
+
+            // TODO: skip null handling for not-nullable primitive inputs after we can completely
+            // trust the `nullable` information.
+            val inputsNullCheck = nullableTypes.zip(inputs)
+              .filter { case (nullable, _) => !nullable }
+              .map { case (_, expr) => IsNull(expr) }
+              .reduceLeftOption[Expression]((e1, e2) => Or(e1, e2))
+            // Once we add an `If` check above the udf, it is safe to mark those checked inputs
+            // as not nullable (i.e., wrap them with `KnownNotNull`), because the null-returning
+            // branch of `If` will be called if any of these checked inputs is null. Thus we can
+            // prevent this rule from being applied repeatedly.
+            val newInputs = nullableTypes.zip(inputs).map { case (nullable, expr) =>
+              if (nullable) expr else KnownNotNull(expr)
+            }
+            inputsNullCheck
+              .map(If(_, Literal.create(null, udf.dataType), udf.copy(children = newInputs)))
+              .getOrElse(udf)
+          }
       }
     }
   }
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/ScalaUDF.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/ScalaUDF.scala
@@ -39,6 +39,7 @@ import org.apache.spark.sql.types.DataType
  * @param nullable  True if the UDF can return null value.
  * @param udfDeterministic  True if the UDF is deterministic. Deterministic UDF returns same result
  *                          each time it is invoked with a particular input.
+ * @param nullableTypes which of the inputTypes are nullable (i.e. not primitive)
  */
 case class ScalaUDF(
     function: AnyRef,
@@ -47,7 +48,8 @@ case class ScalaUDF(
     inputTypes: Seq[DataType] = Nil,
     udfName: Option[String] = None,
     nullable: Boolean = true,
-    udfDeterministic: Boolean = true)
+    udfDeterministic: Boolean = true,
+    nullableTypes: Seq[Boolean] = Nil)
   extends Expression with ImplicitCastInputTypes with NonSQLExpression with UserDefinedExpression {
 
   // The constructor for SPARK 2.1 and 2.2
@@ -58,7 +60,8 @@ case class ScalaUDF(
       inputTypes: Seq[DataType],
       udfName: Option[String]) = {
     this(
-      function, dataType, children, inputTypes, udfName, nullable = true, udfDeterministic = true)
+      function, dataType, children, inputTypes, udfName, nullable = true,
+      udfDeterministic = true, nullableTypes = Nil)
   }
 
   override lazy val deterministic: Boolean = udfDeterministic && children.forall(_.deterministic)
diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/ScalaReflectionSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/ScalaReflectionSuite.scala
@@ -261,23 +261,6 @@ class ScalaReflectionSuite extends SparkFunSuite {
     }
   }
 
-  test("get parameter type from a function object") {
-    val primitiveFunc = (i: Int, j: Long) => "x"
-    val primitiveTypes = getParameterTypes(primitiveFunc)
-    assert(primitiveTypes.forall(_.isPrimitive))
-    assert(primitiveTypes === Seq(classOf[Int], classOf[Long]))
-
-    val boxedFunc = (i: java.lang.Integer, j: java.lang.Long) => "x"
-    val boxedTypes = getParameterTypes(boxedFunc)
-    assert(boxedTypes.forall(!_.isPrimitive))
-    assert(boxedTypes === Seq(classOf[java.lang.Integer], classOf[java.lang.Long]))
-
-    val anyFunc = (i: Any, j: AnyRef) => "x"
-    val anyTypes = getParameterTypes(anyFunc)
-    assert(anyTypes.forall(!_.isPrimitive))
-    assert(anyTypes === Seq(classOf[java.lang.Object], classOf[java.lang.Object]))
-  }
-
   test("SPARK-15062: Get correct serializer for List[_]") {
     val list = List(1, 2, 3)
     val serializer = serializerFor[List[Int]](BoundReference(
diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/analysis/AnalysisSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/analysis/AnalysisSuite.scala
@@ -317,13 +317,15 @@ class AnalysisSuite extends AnalysisTest with Matchers {
     checkUDF(udf1, expected1)
 
     // only primitive parameter needs special null handling
-    val udf2 = ScalaUDF((s: String, d: Double) => "x", StringType, string :: double :: Nil)
+    val udf2 = ScalaUDF((s: String, d: Double) => "x", StringType, string :: double :: Nil,
+      nullableTypes = true :: false :: Nil)
     val expected2 =
       If(IsNull(double), nullResult, udf2.copy(children = string :: KnownNotNull(double) :: Nil))
     checkUDF(udf2, expected2)
 
     // special null handling should apply to all primitive parameters
-    val udf3 = ScalaUDF((s: Short, d: Double) => "x", StringType, short :: double :: Nil)
+    val udf3 = ScalaUDF((s: Short, d: Double) => "x", StringType, short :: double :: Nil,
+      nullableTypes = false :: false :: Nil)
     val expected3 = If(
       IsNull(short) || IsNull(double),
       nullResult,
@@ -335,7 +337,8 @@ class AnalysisSuite extends AnalysisTest with Matchers {
     val udf4 = ScalaUDF(
       (s: Short, d: Double) => "x",
       StringType,
-      short :: double.withNullability(false) :: Nil)
+      short :: double.withNullability(false) :: Nil,
+      nullableTypes = false :: false :: Nil)
     val expected4 = If(
       IsNull(short),
       nullResult,
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/DataFrameStatFunctions.scala b/sql/core/src/main/scala/org/apache/spark/sql/DataFrameStatFunctions.scala
@@ -24,7 +24,8 @@ import scala.collection.JavaConverters._
 import org.apache.spark.annotation.InterfaceStability
 import org.apache.spark.sql.catalyst.InternalRow
 import org.apache.spark.sql.execution.stat._
-import org.apache.spark.sql.functions.col
+import org.apache.spark.sql.expressions.UserDefinedFunction
+import org.apache.spark.sql.functions.{col, udfInternal}
 import org.apache.spark.sql.types._
 import org.apache.spark.util.sketch.{BloomFilter, CountMinSketch}
 
@@ -375,7 +376,7 @@ final class DataFrameStatFunctions private[sql](df: DataFrame) {
     import org.apache.spark.sql.functions.{rand, udf}
     val c = Column(col)
     val r = rand(seed)
-    val f = udf { (stratum: Any, x: Double) =>
+    val f = udfInternal { (stratum: Any, x: Double) =>
       x < fractions.getOrElse(stratum.asInstanceOf[T], 0.0)
     }
     df.filter(f(c, r))
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/UDFRegistration.scala b/sql/core/src/main/scala/org/apache/spark/sql/UDFRegistration.scala
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/expressions/UserDefinedFunction.scala b/sql/core/src/main/scala/org/apache/spark/sql/expressions/UserDefinedFunction.scala
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/functions.scala b/sql/core/src/main/scala/org/apache/spark/sql/functions.scala

Original file line number	Diff line number	Diff line change
`@@ -239,7 +239,7 @@ private[spark] abstract class BasePythonRunner[IN, OUT](`
`239`	`239`	`}`
`240`	`240`	`// Close ServerSocket on task completion.`
`241`	`241`	`serverSocket.foreach { server =>`
`242`		`- context.addTaskCompletionListener(_ => server.close())`
	`242`	`+ context.addTaskCompletionListener[Unit](_ => server.close())`
`243`	`243`	`}`
`244`	`244`	`val boundPort: Int = serverSocket.map(_.getLocalPort).getOrElse(0)`
`245`	`245`	`if (boundPort == -1) {`
Original file line number	Diff line number	Diff line change
`@@ -211,8 +211,8 @@ abstract class PredictionModel[FeaturesType, M <: PredictionModel[FeaturesType,`
`211`	`211`	`}`
`212`	`212`
`213`	`213`	`protected def transformImpl(dataset: Dataset[_]): DataFrame = {`
`214`		`- val predictUDF = udf { (features: Any) =>`
`215`		`- predict(features.asInstanceOf[FeaturesType])`
	`214`	`+ val predictUDF = udfInternal { features: FeaturesType =>`
	`215`	`+ predict(features)`
`216`	`216`	`}`
`217`	`217`	`dataset.withColumn($(predictionCol), predictUDF(col($(featuresCol))))`
`218`	`218`	`}`
Original file line number	Diff line number	Diff line change
`@@ -287,8 +287,8 @@ class GBTClassificationModel private[ml](`
`287`	`287`
`288`	`288`	`override protected def transformImpl(dataset: Dataset[_]): DataFrame = {`
`289`	`289`	`val bcastModel = dataset.sparkSession.sparkContext.broadcast(this)`
`290`		`- val predictUDF = udf { (features: Any) =>`
`291`		`- bcastModel.value.predict(features.asInstanceOf[Vector])`
	`290`	`+ val predictUDF = udfInternal { features: Vector =>`
	`291`	`+ bcastModel.value.predict(features)`
`292`	`292`	`}`
`293`	`293`	`dataset.withColumn($(predictionCol), predictUDF(col($(featuresCol))))`
`294`	`294`	`}`
Original file line number	Diff line number	Diff line change
`@@ -209,8 +209,8 @@ class RandomForestClassificationModel private[ml] (`
`209`	`209`
`210`	`210`	`override protected def transformImpl(dataset: Dataset[_]): DataFrame = {`
`211`	`211`	`val bcastModel = dataset.sparkSession.sparkContext.broadcast(this)`
`212`		`- val predictUDF = udf { (features: Any) =>`
`213`		`- bcastModel.value.predict(features.asInstanceOf[Vector])`
	`212`	`+ val predictUDF = udfInternal { features: Vector =>`
	`213`	`+ bcastModel.value.predict(features)`
`214`	`214`	`}`
`215`	`215`	`dataset.withColumn($(predictionCol), predictUDF(col($(featuresCol))))`
`216`	`216`	`}`
Original file line number	Diff line number	Diff line change
`@@ -165,7 +165,7 @@ class FeatureHasher(@Since("2.3.0") override val uid: String) extends Transforme`
`165`	`165`	`}`
`166`	`166`	`}`
`167`	`167`
`168`		`- val hashFeatures = udf { row: Row =>`
	`168`	`+ val hashFeatures = udfInternal { row: Row =>`
`169`	`169`	`val map = new OpenHashMap[Int, Double]()`
`170`	`170`	`localInputCols.foreach { colName =>`
`171`	`171`	`val fieldIndex = row.fieldIndex(colName)`
Original file line number	Diff line number	Diff line change
`@@ -140,7 +140,7 @@ class VectorAssembler @Since("1.4.0") (@Since("1.4.0") override val uid: String)`
`140`	`140`	`case VectorAssembler.ERROR_INVALID => (dataset, false)`
`141`	`141`	`}`
`142`	`142`	`// Data transformation.`
`143`		`- val assembleFunc = udf { r: Row =>`
	`143`	`+ val assembleFunc = udfInternal { r: Row =>`
`144`	`144`	`VectorAssembler.assemble(lengths, keepInvalid)(r.toSeq: _*)`
`145`	`145`	`}.asNondeterministic()`
`146`	`146`	`val args = $(inputCols).map { c =>`