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
|
/*
* 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.spark.SparkFunSuite
import org.apache.spark.ml.tree.impl.TreeTests
import org.apache.spark.ml.util.MLTestingUtils
import org.apache.spark.mllib.linalg.Vectors
import org.apache.spark.mllib.regression.LabeledPoint
import org.apache.spark.mllib.tree.{EnsembleTestHelper, GradientBoostedTrees => OldGBT}
import org.apache.spark.mllib.tree.configuration.{Algo => OldAlgo}
import org.apache.spark.mllib.util.MLlibTestSparkContext
import org.apache.spark.rdd.RDD
import org.apache.spark.sql.DataFrame
import org.apache.spark.util.Utils
/**
* Test suite for [[GBTRegressor]].
*/
class GBTRegressorSuite extends SparkFunSuite with MLlibTestSparkContext {
import GBTRegressorSuite.compareAPIs
// Combinations for estimators, learning rates and subsamplingRate
private val testCombinations =
Array((10, 1.0, 1.0), (10, 0.1, 1.0), (10, 0.5, 0.75), (10, 0.1, 0.75))
private var data: RDD[LabeledPoint] = _
private var trainData: RDD[LabeledPoint] = _
private var validationData: RDD[LabeledPoint] = _
override def beforeAll() {
super.beforeAll()
data = sc.parallelize(EnsembleTestHelper.generateOrderedLabeledPoints(numFeatures = 10, 100), 2)
trainData =
sc.parallelize(EnsembleTestHelper.generateOrderedLabeledPoints(numFeatures = 20, 120), 2)
validationData =
sc.parallelize(EnsembleTestHelper.generateOrderedLabeledPoints(numFeatures = 20, 80), 2)
}
test("Regression with continuous features") {
val categoricalFeatures = Map.empty[Int, Int]
GBTRegressor.supportedLossTypes.foreach { loss =>
testCombinations.foreach {
case (maxIter, learningRate, subsamplingRate) =>
val gbt = new GBTRegressor()
.setMaxDepth(2)
.setSubsamplingRate(subsamplingRate)
.setLossType(loss)
.setMaxIter(maxIter)
.setStepSize(learningRate)
.setSeed(123)
compareAPIs(data, None, gbt, categoricalFeatures)
}
}
}
test("GBTRegressor behaves reasonably on toy data") {
val df = sqlContext.createDataFrame(Seq(
LabeledPoint(10, Vectors.dense(1, 2, 3, 4)),
LabeledPoint(-5, Vectors.dense(6, 3, 2, 1)),
LabeledPoint(11, Vectors.dense(2, 2, 3, 4)),
LabeledPoint(-6, Vectors.dense(6, 4, 2, 1)),
LabeledPoint(9, Vectors.dense(1, 2, 6, 4)),
LabeledPoint(-4, Vectors.dense(6, 3, 2, 2))
))
val gbt = new GBTRegressor()
.setMaxDepth(2)
.setMaxIter(2)
val model = gbt.fit(df)
// copied model must have the same parent.
MLTestingUtils.checkCopy(model)
val preds = model.transform(df)
val predictions = preds.select("prediction").rdd.map(_.getDouble(0))
// Checks based on SPARK-8736 (to ensure it is not doing classification)
assert(predictions.max() > 2)
assert(predictions.min() < -1)
}
test("Checkpointing") {
val tempDir = Utils.createTempDir()
val path = tempDir.toURI.toString
sc.setCheckpointDir(path)
val df = sqlContext.createDataFrame(data)
val gbt = new GBTRegressor()
.setMaxDepth(2)
.setMaxIter(5)
.setStepSize(0.1)
.setCheckpointInterval(2)
.setSeed(123)
val model = gbt.fit(df)
sc.checkpointDir = None
Utils.deleteRecursively(tempDir)
}
test("should support all NumericType labels and not support other types") {
val gbt = new GBTRegressor().setMaxDepth(1)
MLTestingUtils.checkNumericTypes[GBTRegressionModel, GBTRegressor](
gbt, isClassification = false, sqlContext) { (expected, actual) =>
TreeTests.checkEqual(expected, actual)
}
}
// TODO: Reinstate test once runWithValidation is implemented SPARK-7132
/*
test("runWithValidation stops early and performs better on a validation dataset") {
val categoricalFeatures = Map.empty[Int, Int]
// Set maxIter large enough so that it stops early.
val maxIter = 20
GBTRegressor.supportedLossTypes.foreach { loss =>
val gbt = new GBTRegressor()
.setMaxIter(maxIter)
.setMaxDepth(2)
.setLossType(loss)
.setValidationTol(0.0)
compareAPIs(trainData, None, gbt, categoricalFeatures)
compareAPIs(trainData, Some(validationData), gbt, categoricalFeatures)
}
}
*/
/////////////////////////////////////////////////////////////////////////////
// Tests of feature importance
/////////////////////////////////////////////////////////////////////////////
test("Feature importance with toy data") {
val gbt = new GBTRegressor()
.setMaxDepth(3)
.setMaxIter(5)
.setSubsamplingRate(1.0)
.setStepSize(0.5)
.setSeed(123)
// In this data, feature 1 is very important.
val data: RDD[LabeledPoint] = TreeTests.featureImportanceData(sc)
val categoricalFeatures = Map.empty[Int, Int]
val df: DataFrame = TreeTests.setMetadata(data, categoricalFeatures, 0)
val importances = gbt.fit(df).featureImportances
val mostImportantFeature = importances.argmax
assert(mostImportantFeature === 1)
assert(importances.toArray.sum === 1.0)
assert(importances.toArray.forall(_ >= 0.0))
}
/////////////////////////////////////////////////////////////////////////////
// Tests of model save/load
/////////////////////////////////////////////////////////////////////////////
// TODO: Reinstate test once save/load are implemented SPARK-6725
/*
test("model save/load") {
val tempDir = Utils.createTempDir()
val path = tempDir.toURI.toString
val trees = Range(0, 3).map(_ => OldDecisionTreeSuite.createModel(OldAlgo.Regression)).toArray
val treeWeights = Array(0.1, 0.3, 1.1)
val oldModel = new OldGBTModel(OldAlgo.Regression, trees, treeWeights)
val newModel = GBTRegressionModel.fromOld(oldModel)
// Save model, load it back, and compare.
try {
newModel.save(sc, path)
val sameNewModel = GBTRegressionModel.load(sc, path)
TreeTests.checkEqual(newModel, sameNewModel)
} finally {
Utils.deleteRecursively(tempDir)
}
}
*/
}
private object GBTRegressorSuite extends SparkFunSuite {
/**
* Train 2 models on the given dataset, one using the old API and one using the new API.
* Convert the old model to the new format, compare them, and fail if they are not exactly equal.
*/
def compareAPIs(
data: RDD[LabeledPoint],
validationData: Option[RDD[LabeledPoint]],
gbt: GBTRegressor,
categoricalFeatures: Map[Int, Int]): Unit = {
val numFeatures = data.first().features.size
val oldBoostingStrategy = gbt.getOldBoostingStrategy(categoricalFeatures, OldAlgo.Regression)
val oldGBT = new OldGBT(oldBoostingStrategy, gbt.getSeed.toInt)
val oldModel = oldGBT.run(data)
val newData: DataFrame = TreeTests.setMetadata(data, categoricalFeatures, numClasses = 0)
val newModel = gbt.fit(newData)
// Use parent from newTree since this is not checked anyways.
val oldModelAsNew = GBTRegressionModel.fromOld(
oldModel, newModel.parent.asInstanceOf[GBTRegressor], categoricalFeatures, numFeatures)
TreeTests.checkEqual(oldModelAsNew, newModel)
assert(newModel.numFeatures === numFeatures)
assert(oldModelAsNew.numFeatures === numFeatures)
}
}
|
