Github user yinxusen commented on a diff in the pull request:
https://github.com/apache/spark/pull/7842#discussion_r43213272
--- Diff:
mllib/src/main/scala/org/apache/spark/mllib/pmml/export/PMMLTreeModelUtils.scala
---
@@ -0,0 +1,282 @@
+/*
+ * 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.mllib.pmml.export
+
+import scala.collection.mutable
+import scala.collection.JavaConverters._
+import scala.collection.mutable.MutableList
+
+import org.dmg.pmml.{Node => PMMLNode, Value => PMMLValue, _}
+
+import org.apache.spark.mllib.tree.configuration.{Algo, FeatureType}
+import org.apache.spark.mllib.tree.configuration.Algo._
+import org.apache.spark.mllib.tree.model.{DecisionTreeModel, Node}
+
+private[mllib] object PMMLTreeModelUtils {
+
+ val FieldNamePrefix = "field_"
+
+ def toPMMLTree(dtModel: DecisionTreeModel, modelName: String):
(TreeModel, List[DataField]) = {
+
+ val miningFunctionType = getPMMLMiningFunctionType(dtModel.algo)
+
+ val treeModel = new org.dmg.pmml.TreeModel()
+ .withModelName(modelName)
+ .withFunctionName(miningFunctionType)
+ .withSplitCharacteristic(TreeModel.SplitCharacteristic.BINARY_SPLIT)
+
+ var (rootNode, miningFields, dataFields, classes) =
buildStub(dtModel.topNode, dtModel.algo)
+
+ // adding predicted classes for classification and target field for
regression
+ // for completeness
+ dtModel.algo match {
+
+ case Algo.Classification => {
+ miningFields = miningFields :+ new MiningField()
+ .withName(FieldName.create("class"))
+ .withUsageType(FieldUsageType.PREDICTED)
+
+ val dataField = new DataField()
+ .withName(FieldName.create("class"))
+ .withOpType(OpType.CATEGORICAL)
+ .withValues(classes.asJava)
+ .withDataType(DataType.DOUBLE)
+
+ dataFields = dataFields :+ dataField
+ }
+ case Algo.Regression => {
+ val targetField = FieldName.create("target")
+ val dataField = new DataField(targetField, OpType.CONTINUOUS,
DataType.DOUBLE)
+ dataFields = dataFields :+ dataField
+
+ miningFields = miningFields :+ new MiningField()
+ .withName(targetField)
+ .withUsageType(FieldUsageType.TARGET)
+ }
+ }
+
+ val miningSchema = new MiningSchema()
+ .withMiningFields(miningFields.asJava)
+
+ treeModel.withNode(rootNode)
+ .withMiningSchema(miningSchema)
+
+ (treeModel, dataFields)
+
+ }
+
+ /** Build a pmml tree stub given the root mllib node. */
+ private def buildStub(
+ rootDTNode: Node,
+ algo: Algo): (PMMLNode, List[MiningField], List[DataField],
List[PMMLValue]) = {
+
+ val miningFields = MutableList[MiningField]()
+ val dataFields = mutable.HashMap[String, DataField]()
+ val classes = mutable.MutableList[Double]()
+
+ def buildStubInternal(rootNode: Node, predicate: Predicate): PMMLNode
= {
+
+ // get rootPMML node for the MLLib node
+ val rootPMMLNode = createNode(rootNode)
+ rootPMMLNode.withPredicate(predicate)
+
+ var leftPredicate: Predicate = new True()
+ var rightPredicate: Predicate = new True()
+
+ if (rootNode.split.isDefined) {
+ val fieldName = FieldName.create(FieldNamePrefix +
rootNode.split.get.feature)
+ val dataField = getDataField(rootNode, fieldName).get
+
+ if (!dataFields.get(dataField.getName.getValue).isDefined) {
+ dataFields.put(dataField.getName.getValue, dataField)
+ miningFields += new MiningField()
+ .withName(dataField.getName)
+ .withUsageType(FieldUsageType.ACTIVE)
+
+ } else if (dataField.getOpType != OpType.CONTINUOUS.value()) {
+ appendCategories(
+ dataFields.get(dataField.getName.getValue).get,
+ dataField.getValues.asScala.toList)
+ }
+
+ leftPredicate = getPredicate(rootNode, Some(dataField.getName),
true)
+ rightPredicate = getPredicate(rootNode, Some(dataField.getName),
false)
+
+ }
+ // if left node exist, add the node
+ if (rootNode.leftNode.isDefined) {
+ val leftNode = buildStubInternal(rootNode.leftNode.get,
leftPredicate)
+ rootPMMLNode.withNodes(leftNode)
+ }
+ // if right node exist, add the node
+ if (rootNode.rightNode.isDefined) {
+ val rightNode = buildStubInternal(rootNode.rightNode.get,
rightPredicate)
+ rootPMMLNode.withNodes(rightNode)
+ }
+
+ // add to the list of classes
+ if (rootNode.isLeaf && (algo == Algo.Classification)) {
+ classes += rootNode.predict.predict
+ }
+
+ rootPMMLNode
+ }
+
+ val pmmlTreeRootNode = buildStubInternal(rootDTNode, new True())
+
+ val pmmlValues = classes.toList.distinct.map(doubleVal => new
PMMLValue(doubleVal.toString))
+
+ val result = (pmmlTreeRootNode,
+ sortMiningFields(miningFields.toList),
+ sortedDataFields(dataFields.values.toList),
+ pmmlValues)
+
+ result
+
+ }
+
+ private def sortMiningFields(miningFields: List[MiningField]):
List[MiningField] = {
+ miningFields
+ .sortBy { case mField =>
mField.getName.getValue.replace(FieldNamePrefix, "").toInt }
+ }
+
+ private def sortedDataFields(dataFields: List[DataField]):
List[DataField] = {
+ dataFields.sortBy { case dField =>
dField.getName.getValue.replace(FieldNamePrefix, "").toInt }
+ }
+
+ private def appendCategories(dtField: DataField, values:
List[PMMLValue]): DataField = {
+ if (dtField.getOpType == OpType.CATEGORICAL) {
+
+ val existingValues = dtField.getValues.asScala
+ .groupBy { case category => category.getValue }.toMap
+
+ values.foreach(category => {
+ if (!existingValues.get(category.getValue).isDefined) {
+ dtField.withValues(category)
+ }
+ })
+ }
+
+ dtField
+ }
+
+ /** Get pmml Predicate for a given mlLib tree node. */
+ private def getPredicate(node: Node, fieldName: Option[FieldName],
isLeft: Boolean): Predicate = {
+ // compound predicate if classification and categories list length > 0
+
+ if (node.split.isDefined) {
+
+ require(fieldName.isDefined, "fieldName should not be None, it
should be defined.")
+ val field = fieldName.get
+
+ val split = node.split.get
+ val featureType = split.featureType
+
+ featureType match {
+ case FeatureType.Continuous => {
--- End diff --
redundant curly braces
---
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]
