/
DecisionTreeRegressor.scala
332 lines (280 loc) · 12.5 KB
/
DecisionTreeRegressor.scala
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
/*
* 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.regression
import org.apache.hadoop.fs.Path
import org.json4s.{DefaultFormats, JObject}
import org.json4s.JsonDSL._
import org.apache.spark.annotation.Since
import org.apache.spark.ml.linalg.Vector
import org.apache.spark.ml.param.ParamMap
import org.apache.spark.ml.tree._
import org.apache.spark.ml.tree.DecisionTreeModelReadWrite._
import org.apache.spark.ml.tree.impl.RandomForest
import org.apache.spark.ml.util._
import org.apache.spark.ml.util.Instrumentation.instrumented
import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo, Strategy => OldStrategy}
import org.apache.spark.mllib.tree.model.{DecisionTreeModel => OldDecisionTreeModel}
import org.apache.spark.sql.{Column, DataFrame, Dataset}
import org.apache.spark.sql.functions._
import org.apache.spark.sql.types.StructType
/**
* <a href="http://en.wikipedia.org/wiki/Decision_tree_learning">Decision tree</a>
* learning algorithm for regression.
* It supports both continuous and categorical features.
*/
@Since("1.4.0")
class DecisionTreeRegressor @Since("1.4.0") (@Since("1.4.0") override val uid: String)
extends Regressor[Vector, DecisionTreeRegressor, DecisionTreeRegressionModel]
with DecisionTreeRegressorParams with DefaultParamsWritable {
@Since("1.4.0")
def this() = this(Identifiable.randomUID("dtr"))
// Override parameter setters from parent trait for Java API compatibility.
/** @group setParam */
@Since("1.4.0")
def setMaxDepth(value: Int): this.type = set(maxDepth, value)
/** @group setParam */
@Since("1.4.0")
def setMaxBins(value: Int): this.type = set(maxBins, value)
/** @group setParam */
@Since("1.4.0")
def setMinInstancesPerNode(value: Int): this.type = set(minInstancesPerNode, value)
/** @group setParam */
@Since("3.0.0")
def setMinWeightFractionPerNode(value: Double): this.type = set(minWeightFractionPerNode, value)
@Since("1.4.0")
def setMinInfoGain(value: Double): this.type = set(minInfoGain, value)
/** @group expertSetParam */
@Since("1.4.0")
def setMaxMemoryInMB(value: Int): this.type = set(maxMemoryInMB, value)
/** @group expertSetParam */
@Since("1.4.0")
def setCacheNodeIds(value: Boolean): this.type = set(cacheNodeIds, value)
/**
* Specifies how often to checkpoint the cached node IDs.
* E.g. 10 means that the cache will get checkpointed every 10 iterations.
* This is only used if cacheNodeIds is true and if the checkpoint directory is set in
* [[org.apache.spark.SparkContext]].
* Must be at least 1.
* (default = 10)
* @group setParam
*/
@Since("1.4.0")
def setCheckpointInterval(value: Int): this.type = set(checkpointInterval, value)
/** @group setParam */
@Since("1.4.0")
def setImpurity(value: String): this.type = set(impurity, value)
/** @group setParam */
@Since("1.6.0")
def setSeed(value: Long): this.type = set(seed, value)
/** @group setParam */
@Since("2.0.0")
def setVarianceCol(value: String): this.type = set(varianceCol, value)
/**
* Sets the value of param [[weightCol]].
* If this is not set or empty, we treat all instance weights as 1.0.
* Default is not set, so all instances have weight one.
*
* @group setParam
*/
@Since("3.0.0")
def setWeightCol(value: String): this.type = set(weightCol, value)
override protected def train(
dataset: Dataset[_]): DecisionTreeRegressionModel = instrumented { instr =>
val categoricalFeatures: Map[Int, Int] =
MetadataUtils.getCategoricalFeatures(dataset.schema($(featuresCol)))
val instances = extractInstances(dataset)
val strategy = getOldStrategy(categoricalFeatures)
require(!strategy.bootstrap, "DecisionTreeRegressor does not need bootstrap sampling")
instr.logPipelineStage(this)
instr.logDataset(instances)
instr.logParams(this, params: _*)
val trees = RandomForest.run(instances, strategy, numTrees = 1, featureSubsetStrategy = "all",
seed = $(seed), instr = Some(instr), parentUID = Some(uid))
trees.head.asInstanceOf[DecisionTreeRegressionModel]
}
/** (private[ml]) Create a Strategy instance to use with the old API. */
private[ml] def getOldStrategy(categoricalFeatures: Map[Int, Int]): OldStrategy = {
super.getOldStrategy(categoricalFeatures, numClasses = 0, OldAlgo.Regression, getOldImpurity,
subsamplingRate = 1.0)
}
@Since("1.4.0")
override def copy(extra: ParamMap): DecisionTreeRegressor = defaultCopy(extra)
}
@Since("1.4.0")
object DecisionTreeRegressor extends DefaultParamsReadable[DecisionTreeRegressor] {
/** Accessor for supported impurities: variance */
final val supportedImpurities: Array[String] = HasVarianceImpurity.supportedImpurities
@Since("2.0.0")
override def load(path: String): DecisionTreeRegressor = super.load(path)
}
/**
* <a href="http://en.wikipedia.org/wiki/Decision_tree_learning">
* Decision tree (Wikipedia)</a> model for regression.
* It supports both continuous and categorical features.
*
* @param rootNode Root of the decision tree
*/
@Since("1.4.0")
class DecisionTreeRegressionModel private[ml] (
override val uid: String,
override val rootNode: Node,
override val numFeatures: Int)
extends RegressionModel[Vector, DecisionTreeRegressionModel]
with DecisionTreeModel with DecisionTreeRegressorParams with MLWritable with Serializable {
/** @group setParam */
def setVarianceCol(value: String): this.type = set(varianceCol, value)
require(rootNode != null,
"DecisionTreeRegressionModel given null rootNode, but it requires a non-null rootNode.")
/**
* Construct a decision tree regression model.
*
* @param rootNode Root node of tree, with other nodes attached.
*/
private[ml] def this(rootNode: Node, numFeatures: Int) =
this(Identifiable.randomUID("dtr"), rootNode, numFeatures)
override def predict(features: Vector): Double = {
rootNode.predictImpl(features).prediction
}
/** We need to update this function if we ever add other impurity measures. */
protected def predictVariance(features: Vector): Double = {
rootNode.predictImpl(features).impurityStats.calculate()
}
@Since("1.4.0")
override def transformSchema(schema: StructType): StructType = {
var outputSchema = super.transformSchema(schema)
if (isDefined(varianceCol) && $(varianceCol).nonEmpty) {
outputSchema = SchemaUtils.updateNumeric(outputSchema, $(varianceCol))
}
if ($(leafCol).nonEmpty) {
outputSchema = SchemaUtils.updateField(outputSchema, getLeafField($(leafCol)))
}
outputSchema
}
@Since("2.0.0")
override def transform(dataset: Dataset[_]): DataFrame = {
val outputSchema = transformSchema(dataset.schema, logging = true)
var predictionColNames = Seq.empty[String]
var predictionColumns = Seq.empty[Column]
if ($(predictionCol).nonEmpty) {
val predictUDF = udf { features: Vector => predict(features) }
predictionColNames :+= $(predictionCol)
predictionColumns :+= predictUDF(col($(featuresCol)))
.as($(predictionCol), outputSchema($(predictionCol)).metadata)
}
if (isDefined(varianceCol) && $(varianceCol).nonEmpty) {
val predictVarianceUDF = udf { features: Vector => predictVariance(features) }
predictionColNames :+= $(varianceCol)
predictionColumns :+= predictVarianceUDF(col($(featuresCol)))
.as($(varianceCol), outputSchema($(varianceCol)).metadata)
}
if ($(leafCol).nonEmpty) {
val leafUDF = udf { features: Vector => predictLeaf(features) }
predictionColNames :+= $(leafCol)
predictionColumns :+= leafUDF(col($(featuresCol)))
.as($(leafCol), outputSchema($(leafCol)).metadata)
}
if (predictionColNames.nonEmpty) {
dataset.withColumns(predictionColNames, predictionColumns)
} else {
this.logWarning(s"$uid: DecisionTreeRegressionModel.transform() does nothing" +
" because no output columns were set.")
dataset.toDF()
}
}
@Since("1.4.0")
override def copy(extra: ParamMap): DecisionTreeRegressionModel = {
copyValues(new DecisionTreeRegressionModel(uid, rootNode, numFeatures), extra).setParent(parent)
}
@Since("1.4.0")
override def toString: String = {
s"DecisionTreeRegressionModel: uid=$uid, depth=$depth, numNodes=$numNodes, " +
s"numFeatures=$numFeatures"
}
/**
* Estimate of the importance of each feature.
*
* This generalizes the idea of "Gini" importance to other losses,
* following the explanation of Gini importance from "Random Forests" documentation
* by Leo Breiman and Adele Cutler, and following the implementation from scikit-learn.
*
* This feature importance is calculated as follows:
* - importance(feature j) = sum (over nodes which split on feature j) of the gain,
* where gain is scaled by the number of instances passing through node
* - Normalize importances for tree to sum to 1.
*
* @note Feature importance for single decision trees can have high variance due to
* correlated predictor variables. Consider using a [[RandomForestRegressor]]
* to determine feature importance instead.
*/
@Since("2.0.0")
lazy val featureImportances: Vector = TreeEnsembleModel.featureImportances(this, numFeatures)
/** Convert to spark.mllib DecisionTreeModel (losing some information) */
override private[spark] def toOld: OldDecisionTreeModel = {
new OldDecisionTreeModel(rootNode.toOld(1), OldAlgo.Regression)
}
@Since("2.0.0")
override def write: MLWriter =
new DecisionTreeRegressionModel.DecisionTreeRegressionModelWriter(this)
}
@Since("2.0.0")
object DecisionTreeRegressionModel extends MLReadable[DecisionTreeRegressionModel] {
@Since("2.0.0")
override def read: MLReader[DecisionTreeRegressionModel] =
new DecisionTreeRegressionModelReader
@Since("2.0.0")
override def load(path: String): DecisionTreeRegressionModel = super.load(path)
private[DecisionTreeRegressionModel]
class DecisionTreeRegressionModelWriter(instance: DecisionTreeRegressionModel)
extends MLWriter {
override protected def saveImpl(path: String): Unit = {
val extraMetadata: JObject = Map(
"numFeatures" -> instance.numFeatures)
DefaultParamsWriter.saveMetadata(instance, path, sc, Some(extraMetadata))
val (nodeData, _) = NodeData.build(instance.rootNode, 0)
val dataPath = new Path(path, "data").toString
val numDataParts = NodeData.inferNumPartitions(instance.numNodes)
sparkSession.createDataFrame(nodeData).repartition(numDataParts).write.parquet(dataPath)
}
}
private class DecisionTreeRegressionModelReader
extends MLReader[DecisionTreeRegressionModel] {
/** Checked against metadata when loading model */
private val className = classOf[DecisionTreeRegressionModel].getName
override def load(path: String): DecisionTreeRegressionModel = {
implicit val format = DefaultFormats
val metadata = DefaultParamsReader.loadMetadata(path, sc, className)
val numFeatures = (metadata.metadata \ "numFeatures").extract[Int]
val root = loadTreeNodes(path, metadata, sparkSession)
val model = new DecisionTreeRegressionModel(metadata.uid, root, numFeatures)
metadata.getAndSetParams(model)
model
}
}
/** Convert a model from the old API */
private[ml] def fromOld(
oldModel: OldDecisionTreeModel,
parent: DecisionTreeRegressor,
categoricalFeatures: Map[Int, Int],
numFeatures: Int = -1): DecisionTreeRegressionModel = {
require(oldModel.algo == OldAlgo.Regression,
s"Cannot convert non-regression DecisionTreeModel (old API) to" +
s" DecisionTreeRegressionModel (new API). Algo is: ${oldModel.algo}")
val rootNode = Node.fromOld(oldModel.topNode, categoricalFeatures)
val uid = if (parent != null) parent.uid else Identifiable.randomUID("dtr")
new DecisionTreeRegressionModel(uid, rootNode, numFeatures)
}
}