Github user selvinsource commented on a diff in the pull request:
https://github.com/apache/spark/pull/7842#discussion_r37694820
--- Diff:
mllib/src/main/scala/org/apache/spark/mllib/pmml/export/TreeModelUtils.scala ---
@@ -0,0 +1,275 @@
+/*
+ * 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.annotation.tailrec
+import scala.collection.JavaConverters._
+import scala.collection.mutable.MutableList
+
+import org.dmg.pmml.{Node => PMMLNode, _}
+
+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 TreeModelUtils {
+
+
+ def toPMMLTree(mllibTreeModel: DecisionTreeModel, modelName: String):
TreeModel = {
+
+ val miningFunctionType = getPMMLMiningFunctionType(mllibTreeModel.algo)
+
+ val miningSchema = new MiningSchema()
+ .withMiningFields(getMiningFieldsForTree(mllibTreeModel).asJava)
+
+ val treeModel = new org.dmg.pmml.TreeModel()
+ .withModelName(modelName)
+ .withFunctionName(miningFunctionType)
+ .withNode(buildStub(mllibTreeModel.topNode))
+ .withMiningSchema(miningSchema)
+
+ treeModel
+
+ }
+
+ /** Build a pmml tree stub given the root mllib node. */
+ private def buildStub(rootNode: Node): PMMLNode = {
+ // get rootPMML node for the MLLib node
+ val rootPMMLNode = createNode(rootNode)
+
+ if (!rootNode.isLeaf) {
+ // if left node exist, add the node
+ if (rootNode.leftNode.isDefined) {
+ val leftNode = buildStub(rootNode.leftNode.get)
+ rootPMMLNode.withNodes(leftNode)
+ }
+ // if right node exist, add the node
+ if (rootNode.rightNode.isDefined) {
+ val rightNode = buildStub(rootNode.rightNode.get)
+ rootPMMLNode.withNodes(rightNode)
+ }
+ }
+
+ rootPMMLNode
+ }
+
+ /** Get pmml Predicate for a given mlLib tree node. */
+ private def getPredicate(node: Node): Option[Predicate] = {
+ // compound predicate if classification and categories list length > 0
+
+ if (node.split.isDefined) {
+ val split = node.split.get
+ val featureType = split.featureType
+ val fieldName = FieldName.create("field_" + split.feature.toString)
+ featureType match {
+ case FeatureType.Continuous => {
+ val value = split.threshold.toString
+ Some(
+ new SimplePredicate(fieldName,
SimplePredicate.Operator.LESS_OR_EQUAL)
+ .withValue(value)
+ )
+ }
+ case FeatureType.Categorical => {
+ if (split.categories.length > 1) {
+ val predicates: List[Predicate] =
+ for (category <- split.categories)
+ yield
+ new SimplePredicate(fieldName,
SimplePredicate.Operator.EQUAL)
+ .withValue(category.toString)
+
+ val compoundPredicate = new CompoundPredicate()
+ .withBooleanOperator(CompoundPredicate.BooleanOperator.OR)
+ .withPredicates(predicates.asJava)
+
+ Some(compoundPredicate)
+
+ }
+ else {
+ val value = split.categories(0).toString
+ Some(new SimplePredicate(fieldName,
SimplePredicate.Operator.EQUAL).withValue(value))
+ }
+ }
+ }
+ }
+ else {
+ None
+ }
+
+ }
+
+ /** Create equivalent PMML node for given mlLib node. */
+ private def createNode(mlLibNode: Node): PMMLNode = {
+ val node = new PMMLNode()
+ .withId(mlLibNode.id.toString)
+ .withScore(mlLibNode.predict.predict.toString)
+
+ val predicate = getPredicate(mlLibNode)
+ if (predicate.isDefined) {
+ node.withPredicate(predicate.get)
+ }
+
+ node
+ }
+
+ /** Get PMML mining function type for given mlLib Algo. */
+ private def getPMMLMiningFunctionType(mlLibAlgo: Algo):
MiningFunctionType = {
+ mlLibAlgo match {
+ case Algo.Classification => MiningFunctionType.CLASSIFICATION
+ case Algo.Regression => MiningFunctionType.REGRESSION
+ }
+ }
+
+ /** Get PMML datafield based on the mllib split feature. */
+ private def getDataField(mllibNode: Node): Option[DataField] = {
+ if (!mllibNode.isLeaf && mllibNode.split.isDefined) {
+ val split = mllibNode.split.get
+ val dataField = new DataField()
+ .withName(FieldName.create("field_" + split.feature.toString))
+
.withDataType(DataType.fromValue(split.threshold.getClass.getSimpleName.toLowerCase))
+
.withOpType(OpType.fromValue(split.featureType.toString.toLowerCase))
+
+ split.featureType match {
+ case FeatureType.Continuous =>
dataField.withOpType(OpType.CONTINUOUS)
+ case FeatureType.Categorical => {
+ dataField.withOpType(OpType.CATEGORICAL)
+ val categories = split.categories
+ .map(category => new
org.dmg.pmml.Value(category.toString)).asJava
+ dataField.withValues(categories)
+ }
+ }
+
+ Some(dataField)
+ }
+ else {
+ None
+ }
+ }
+
+ /** Get PMML Mining field based on Mllib node split feature. */
+ private def getMiningField(mllibNode: Node): Option[MiningField] = {
+ if (!mllibNode.isLeaf && mllibNode.split.isDefined) {
+ val split = mllibNode.split.get
+
+ val miningField = new MiningField()
+ .withName(FieldName.create("field_" + split.feature.toString))
+ .withUsageType(FieldUsageType.ACTIVE)
+
+ Some(miningField)
+ }
+ else {
+ None
+ }
+
+ }
+
+ /** Get distinct PMML mining fields list for a given mlLib decision tree
model. */
+ def getMiningFieldsForTree(treeModel: DecisionTreeModel):
List[MiningField] = {
+
+ @tailrec
+ def getMiningFieldsForTreeRec(
+ nodeList: List[Node],
+ miningFlds: List[MiningField]):
List[MiningField] = {
+
+ nodeList match {
+ case Nil => miningFlds
+ case nd :: ls if (nd.isLeaf) => miningFlds
+ case nd :: ls if (!nd.isLeaf && nd.split.isDefined) => {
+ val ndList = MutableList[Node]()
+ if (nd.leftNode.isDefined) {
+ ndList += nd.leftNode.get
+ }
+
+ if (nd.rightNode.isDefined) {
+ ndList += nd.rightNode.get
+ }
+
+ getMiningFieldsForTreeRec(ndList.toList, miningFlds :+
getMiningField(nd).get)
+
+ }
+ }
+ }
+
+ val miningFields = getMiningFieldsForTreeRec(List(treeModel.topNode),
List[MiningField]())
--- End diff --
getMiningFieldsForTreeRec is very similar to getDataFieldsForTreeRec,
couldn't you merge these?
Also, I am not sure you even need getMiningFieldsForTreeRec, you could
create the mining fields from the data fields you already have, usually (in the
models I exported) the field names were the same for mining fields and data
fields.
---
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]
