Github user jkbradley commented on a diff in the pull request: https://github.com/apache/spark/pull/12023#discussion_r57674671 --- Diff: mllib/src/main/scala/org/apache/spark/ml/tree/treeModels.scala --- @@ -238,3 +238,128 @@ private[ml] object DecisionTreeModelReadWrite { finalNodes.head } } + +private[ml] object RandomForestModelReadWrite { + + /** + * Info for a [[org.apache.spark.ml.tree.Split]] + * + * @param featureIndex Index of feature split on + * @param leftCategoriesOrThreshold For categorical feature, set of leftCategories. + * For continuous feature, threshold. + * @param numCategories For categorical feature, number of categories. + * For continuous feature, -1. + */ + case class SplitData( + featureIndex: Int, + leftCategoriesOrThreshold: Array[Double], + numCategories: Int) { + + def getSplit: Split = { + if (numCategories != -1) { + new CategoricalSplit(featureIndex, leftCategoriesOrThreshold, numCategories) + } else { + assert(leftCategoriesOrThreshold.length == 1, s"DecisionTree split data expected" + + s" 1 threshold for ContinuousSplit, but found thresholds: " + + leftCategoriesOrThreshold.mkString(", ")) + new ContinuousSplit(featureIndex, leftCategoriesOrThreshold(0)) + } + } + } + + object SplitData { + def apply(split: Split): SplitData = split match { + case s: CategoricalSplit => + SplitData(s.featureIndex, s.leftCategories, s.numCategories) + case s: ContinuousSplit => + SplitData(s.featureIndex, Array(s.threshold), -1) + } + } + + /** + * Info for a [[Node]] + * + * @param treeID Index used for tree identification in RandomForest + * @param id Index used for tree reconstruction. Indices follow a pre-order traversal. + * @param impurityStats Stats array. Impurity type is stored in metadata. + * @param gain Gain, or arbitrary value if leaf node. + * @param leftChild Left child index, or arbitrary value if leaf node. + * @param rightChild Right child index, or arbitrary value if leaf node. + * @param split Split info, or arbitrary value if leaf node. + */ + case class NodeData( + treeID: Int, + id: Int, + prediction: Double, + impurity: Double, + impurityStats: Array[Double], + gain: Double, + leftChild: Int, + rightChild: Int, + split: SplitData) + + object NodeData { + /** + * Create [[NodeData]] instances for this node and all children. + * + * @param id Current ID. IDs are assigned via a pre-order traversal. + * @return (sequence of nodes in pre-order traversal order, largest ID in subtree) + * The nodes are returned in pre-order traversal (root first) so that it is easy to + * get the ID of the subtree's root node. + */ + def build(node: Node, treeID: Int, id: Int): (Seq[NodeData], Int) = node match { + case n: InternalNode => + val (leftNodeData, leftIdx) = build(n.leftChild, treeID, id + 1) + val (rightNodeData, rightIdx) = build(n.rightChild, treeID, leftIdx + 1) + val thisNodeData = NodeData(treeID, id, n.prediction, n.impurity, n.impurityStats.stats, + n.gain, leftNodeData.head.id, rightNodeData.head.id, SplitData(n.split)) + (thisNodeData +: (leftNodeData ++ rightNodeData), rightIdx) + case _: LeafNode => + (Seq(NodeData(treeID, id, node.prediction, node.impurity, node.impurityStats.stats, + -1.0, -1, -1, SplitData(-1, Array.empty[Double], -1))), + id) + } + } + + def loadTreeNodes( --- End diff -- This should not be copied either. Please reuse the one for trees.
--- 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