/*
* 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.mahout.classifier.sgd;
import org.apache.mahout.common.MahoutTestCase;
import org.apache.mahout.common.RandomUtils;
import org.apache.mahout.math.DenseVector;
import org.apache.mahout.math.Vector;
import org.apache.mahout.math.jet.random.Exponential;
import org.junit.Test;
import java.util.Random;
public final class AdaptiveLogisticRegressionTest extends MahoutTestCase {
@Test
public void testTrain() {
Random gen = RandomUtils.getRandom();
Exponential exp = new Exponential(0.5, gen);
Vector beta = new DenseVector(200);
for (Vector.Element element : beta.all()) {
int sign = 1;
if (gen.nextDouble() < 0.5) {
sign = -1;
}
element.set(sign * exp.nextDouble());
}
AdaptiveLogisticRegression.Wrapper cl = new AdaptiveLogisticRegression.Wrapper(2, 200, new L1());
cl.update(new double[]{1.0e-5, 1});
for (int i = 0; i < 10000; i++) {
AdaptiveLogisticRegression.TrainingExample r = getExample(i, gen, beta);
cl.train(r);
if (i % 1000 == 0) {
// cl.close();
System.out.printf("%10d %10.3f\n", i, cl.getLearner().auc());
}
}
assertEquals(1, cl.getLearner().auc(), 0.1);
AdaptiveLogisticRegression x = new AdaptiveLogisticRegression(2, 200, new L1());
x.setInterval(1000);
for (int i = 0; i < 20000; i++) {
AdaptiveLogisticRegression.TrainingExample r = getExample(i, gen, beta);
x.train(r.getKey(), r.getActual(), r.getInstance());
if (i % 1000 == 0 && x.getBest() != null) {
System.out.printf("%10d %10.4f %10.8f %.3f\n",
i, x.auc(),
Math.log10(x.getBest().getMappedParams()[0]), x.getBest().getMappedParams()[1]);
}
}
assertEquals(1, x.auc(), 0.1);
}
private static AdaptiveLogisticRegression.TrainingExample getExample(int i, Random gen, Vector beta) {
Vector data = new DenseVector(200);
for (Vector.Element element : data.all()) {
element.set(gen.nextDouble() < 0.3 ? 1 : 0);
}
double p = 1 / (1 + Math.exp(1.5 - data.dot(beta)));
int target = 0;
if (gen.nextDouble() < p) {
target = 1;
}
return new AdaptiveLogisticRegression.TrainingExample(i, null, target, data);
}
@Test
public void copyLearnsAsExpected() {
Random gen = RandomUtils.getRandom();
Exponential exp = new Exponential(0.5, gen);
Vector beta = new DenseVector(200);
for (Vector.Element element : beta.all()) {
int sign = 1;
if (gen.nextDouble() < 0.5) {
sign = -1;
}
element.set(sign * exp.nextDouble());
}
// train one copy of a wrapped learner
AdaptiveLogisticRegression.Wrapper w = new AdaptiveLogisticRegression.Wrapper(2, 200, new L1());
for (int i = 0; i < 3000; i++) {
AdaptiveLogisticRegression.TrainingExample r = getExample(i, gen, beta);
w.train(r);
if (i % 1000 == 0) {
System.out.printf("%10d %.3f\n", i, w.getLearner().auc());
}
}
System.out.printf("%10d %.3f\n", 3000, w.getLearner().auc());
double auc1 = w.getLearner().auc();
// then switch to a copy of that learner ... progress should continue
AdaptiveLogisticRegression.Wrapper w2 = w.copy();
for (int i = 0; i < 5000; i++) {
if (i % 1000 == 0) {
if (i == 0) {
assertEquals("Should have started with no data", 0.5, w2.getLearner().auc(), 0.0001);
}
if (i == 1000) {
double auc2 = w2.getLearner().auc();
assertTrue("Should have had head-start", Math.abs(auc2 - 0.5) > 0.1);
assertTrue("AUC should improve quickly on copy", auc1 < auc2);
}
System.out.printf("%10d %.3f\n", i, w2.getLearner().auc());
}
AdaptiveLogisticRegression.TrainingExample r = getExample(i, gen, beta);
w2.train(r);
}
assertEquals("Original should not change after copy is updated", auc1, w.getLearner().auc(), 1.0e-5);
// this improvement is really quite lenient
assertTrue("AUC should improve significantly on copy", auc1 < w2.getLearner().auc() - 0.05);
// make sure that the copy didn't lose anything
assertEquals(auc1, w.getLearner().auc(), 0);
}
@Test
public void stepSize() {
assertEquals(500, AdaptiveLogisticRegression.stepSize(15000, 2));
assertEquals(2000, AdaptiveLogisticRegression.stepSize(15000, 2.6));
assertEquals(5000, AdaptiveLogisticRegression.stepSize(24000, 2.6));
assertEquals(10000, AdaptiveLogisticRegression.stepSize(15000, 3));
}
@Test
public void constantStep() {
AdaptiveLogisticRegression lr = new AdaptiveLogisticRegression(2, 1000, new L1());
lr.setInterval(5000);
assertEquals(20000, lr.nextStep(15000));
assertEquals(20000, lr.nextStep(15001));
assertEquals(20000, lr.nextStep(16500));
assertEquals(20000, lr.nextStep(19999));
}
@Test
public void growingStep() {
AdaptiveLogisticRegression lr = new AdaptiveLogisticRegression(2, 1000, new L1());
lr.setInterval(2000, 10000);
// start with minimum step size
for (int i = 2000; i < 20000; i+=2000) {
assertEquals(i + 2000, lr.nextStep(i));
}
// then level up a bit
for (int i = 20000; i < 50000; i += 5000) {
assertEquals(i + 5000, lr.nextStep(i));
}
// and more, but we top out with this step size
for (int i = 50000; i < 500000; i += 10000) {
assertEquals(i + 10000, lr.nextStep(i));
}
}
}