Github user hhbyyh commented on a diff in the pull request:
https://github.com/apache/spark/pull/17583#discussion_r121312459
--- Diff:
mllib/src/main/scala/org/apache/spark/ml/feature/FuncTransformer.scala ---
@@ -0,0 +1,150 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements. See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.ml.feature
+
+import java.io.{ByteArrayInputStream, ByteArrayOutputStream,
ObjectInputStream, ObjectOutputStream}
+
+import scala.reflect.runtime.universe.TypeTag
+
+import org.apache.hadoop.fs.Path
+
+import org.apache.spark.annotation.{DeveloperApi, Since}
+import org.apache.spark.ml.UnaryTransformer
+import org.apache.spark.ml.feature.FuncTransformer.FuncTransformerWriter
+import org.apache.spark.ml.param.ParamMap
+import org.apache.spark.ml.util._
+import org.apache.spark.sql.Row
+import org.apache.spark.sql.catalyst.ScalaReflection
+import org.apache.spark.sql.types.DataType
+
+/**
+ * :: DeveloperApi ::
+ * FuncTransformer allows easily creation of a custom feature transformer
for DataFrame, such like
+ * conditional conversion(if...else...), type conversion, array indexing
and many string ops.
+ * Note that FuncTransformer supports serialization via scala
ObjectOutputStream and may not
+ * guarantee save/load compatibility between different scala version.
+ */
+@DeveloperApi
+@Since("2.3.0")
+class FuncTransformer [IN: TypeTag, OUT: TypeTag] @Since("2.3.0") (
+ @Since("2.3.0") override val uid: String,
+ @Since("2.3.0") val func: IN => OUT,
+ @Since("2.3.0") val outputDataType: DataType
+ ) extends UnaryTransformer[IN, OUT, FuncTransformer[IN, OUT]] with
DefaultParamsWritable {
+
+ /**
+ * Create a FuncTransformer with specific function and output data type.
+ * @param fx function which converts an input object to output object.
+ * @param outputDataType specific output data type
+ */
+ @Since("2.3.0")
+ def this(fx: IN => OUT, outputDataType: DataType) =
+ this(Identifiable.randomUID("FuncTransformer"), fx, outputDataType)
+
+ /**
+ * Create a FuncTransformer with specific function and automatically
infer the output data type.
+ * If the output data type cannot be automatically inferred, an
exception will be thrown.
+ * @param fx function which converts an input object to output object.
+ */
+ @Since("2.3.0")
+ def this(fx: IN => OUT) =
this(Identifiable.randomUID("FuncTransformer"), fx,
+ try {
+ ScalaReflection.schemaFor[OUT].dataType
+ } catch {
+ case _: UnsupportedOperationException => throw new
UnsupportedOperationException(
+ s"FuncTransformer outputDataType cannot be automatically inferred,
please try" +
+ s" the constructor with specific outputDataType")
+ }
+ )
+
+ setDefault(inputCol -> "input", outputCol -> "output")
+
+ @Since("2.3.0")
+ override def createTransformFunc: IN => OUT = func
+
+ @Since("2.3.0")
+ override def write: MLWriter = new FuncTransformerWriter(
+ this.asInstanceOf[FuncTransformer[Nothing, Nothing]])
+
+ @Since("2.3.0")
+ override def copy(extra: ParamMap): FuncTransformer[IN, OUT] = {
+ copyValues(new FuncTransformer(uid, func, outputDataType), extra)
+ }
+
+ override protected def validateInputType(inputType: DataType): Unit = {
+ try {
+ val funcINType = ScalaReflection.schemaFor[IN].dataType
+ require(inputType.equals(funcINType),
+ s"$uid only accept input type $funcINType but got $inputType.")
+ } catch {
+ case _: UnsupportedOperationException =>
+ // cannot infer the output data type, log warning but do not block
transform
+ logWarning(s"FuncTransformer input Type cannot be automatically
inferred," +
+ s"Type check omitted for $uid")
+ }
+ }
+}
+
+/**
+ * :: DeveloperApi ::
+ * Companion object for FuncTransformer with save and load function.
+ */
+@DeveloperApi
+@Since("2.3.0")
+object FuncTransformer extends
DefaultParamsReadable[FuncTransformer[Nothing, Nothing]] {
+
+ private[FuncTransformer]
+ class FuncTransformerWriter(instance: FuncTransformer[Nothing, Nothing])
extends MLWriter {
+
+ private case class Data(func: Array[Byte], dataType: String)
+
+ override protected def saveImpl(path: String): Unit = {
+ DefaultParamsWriter.saveMetadata(instance, path, sc)
+ val bo = new ByteArrayOutputStream()
+ new ObjectOutputStream(bo).writeObject(instance.func)
+ val data = Data(bo.toByteArray, instance.outputDataType.json)
+ val dataPath = new Path(path, "data").toString
+
sparkSession.createDataFrame(Seq(data)).repartition(1).write.parquet(dataPath)
+ }
+ }
+
+ private class FuncTransformerReader extends
MLReader[FuncTransformer[Nothing, Nothing]] {
+
+ private val className = classOf[FuncTransformer[Nothing,
Nothing]].getName
+
+ override def load(path: String): FuncTransformer[Nothing, Nothing] = {
+ val metadata = DefaultParamsReader.loadMetadata(path, sc, className)
+ val dataPath = new Path(path, "data").toString
+ val data = sparkSession.read.parquet(dataPath)
+ val Row(funcBytes: Array[Byte], dataType: String) = data
+ .select("func", "dataType")
+ .head()
+ val func = new ObjectInputStream(new
ByteArrayInputStream(funcBytes)).readObject()
+ val model = new FuncTransformer(
+ metadata.uid, func.asInstanceOf[Function[Any, Any]],
DataType.fromJson(dataType))
+ DefaultParamsReader.getAndSetParams(model, metadata)
+ model.asInstanceOf[FuncTransformer[Nothing, Nothing]]
--- End diff --
After loading, the `FuncTransformer[Nothing, Nothing]` will cause warning
during `validateInputType`. The warning will not block `transform`, but it
would be better if we can fully restore the `FuncTransformer` with original
Type.
I'll keep checking, but any suggestion is welcome.
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
