@Test
public void testProductResult() {
int tS = (int) Math.pow(2, 12);
int tSum = tS * (tS + 1) / 2;
for (short k = 0; k < tGP12.getBladeCount(); k++) {
short[] tSpot = tGP12.getResult(k);
int tSumP = 0;
for (int j = 0; j < tSpot.length; j++) {
tSumP += Math.abs(tSpot[j]);
}
assertTrue(tSum == tSumP);
}
tS = (int) Math.pow(2, 8);
tSum = tS * (tS + 1) / 2;
for (short k = 0; k < tGP8.getBladeCount(); k++) {
short[] tSpot = tGP8.getResult(k);
int tSumP = 0;
for (int j = 0; j < tSpot.length; j++) {
tSumP += Math.abs(tSpot[j]);
}
assertTrue(tSum == tSumP);
}
tS = (int) Math.pow(2, 4);
tSum = tS * (tS + 1) / 2;
for (short k = 0; k < tGP4.getBladeCount(); k++) {
short[] tSpot = tGP4.getResult(k);
int tSumP = 0;
for (int j = 0; j < tSpot.length; j++) {
tSumP += Math.abs(tSpot[j]);
}
assertTrue(tSum == tSumP);
}
}
protected boolean sufficientlyEqual(Object value1, Object value2) {
if (value1 == null) return value2 == null;
if (value1 instanceof Date && value2 instanceof Date) {
// for dynamic dates.
return Math.abs(((Date) value1).getTime() - ((Date) value2).getTime()) < 200L;
} else {
return value1.equals(value2);
}
}
@Test
public void testConstructors() {
ValueGene test = new DoubleValueGene(93947810231.0);
Assert.assertTrue(
"value constructor failed",
Math.abs(test.getValue().getNumber().doubleValue() - 93947810231.0) < 1000);
test = new DoubleValueGene(new MutableDouble(20237420342.0));
Assert.assertTrue(
"MutableDouble value constructor failed",
Math.abs(test.getValue().getNumber().doubleValue() - 20237420342.0) < 1000);
test = new DoubleValueGene(82649173937.0);
Assert.assertTrue(
"Number value constructor failed",
Math.abs(test.getValue().getNumber().doubleValue() - 82649173937.0) < 1000);
test = new DoubleValueGene();
Assert.assertTrue(
"default constructor failed", test.getValue().getNumber().doubleValue() == 0.0);
}
@Test
public void testscaleValues() {
LinkedList<Double> variables = new LinkedList<Double>();
double scale = 2;
for (double d : p.getVariables()) variables.add(d);
p.scale(scale);
boolean equals = (p.getVariables().size() == variables.size());
for (int i = 0; i < p.getVariables().size(); i++)
equals =
equals
&& (Math.abs(p.getVariables().get(i) - variables.get(i))
< InfinitePoint.getEpsilon());
}
@Test
public void testNoise() {
System.out.println("test noise generator");
int iters = 100000;
double mean = 0.2;
double var = 1.0 / 11.0;
CircularRandomVariable instance = new WrappedUniform(mean, var);
double umean = instance.intrinsicMean();
assertEquals(-0.3, umean, 0.000001);
double sum = 0;
double sum2 = 0;
for (int i = 0; i <= iters; i++) {
double noise = instance.noise();
boolean result = noise >= -0.5 && noise < 0.5;
sum2 += fracpart(noise - umean) * fracpart(noise - umean);
assertEquals(true, result);
}
// test the unwrapped variance
assertTrue(Math.abs(sum2 / iters - instance.intrinsicVariance()) < 0.01);
}
@Test
public void testCovariance() {
DataFrame df = context.table("testData2");
Double result = df.stat().cov("a", "b");
Assert.assertTrue(Math.abs(result) < 1.0e-6);
}
@Test
public void testCorrelation() {
DataFrame df = context.table("testData2");
Double pearsonCorr = df.stat().corr("a", "b", "pearson");
Assert.assertTrue(Math.abs(pearsonCorr) < 1.0e-6);
}
/**
* Test to make sure that chunking does not change the results.
*
* @throws ClassNotFoundException
* @throws IllegalAccessException
* @throws InstantiationException
*/
@Test
public void testNoCache()
throws InstantiationException, IllegalAccessException, ClassNotFoundException {
ProblemSet ps =
new ProblemSet(
Paths.get(JSANConstants.JSAN_PROBLEMSETS_PREFIX, "drexel_1_small.xml").toString());
Document d = ps.trainDocAt("a", "a_01.txt");
Assert.assertNotNull("No document a_01.txt found.", d);
ps.removeTrainDocAt("a", d);
ps.addTestDoc("a", d);
Path path =
Paths.get(JSANConstants.JSAN_FEATURESETS_PREFIX, "writeprints_feature_set_limited.xml");
FullAPI test =
new FullAPI.Builder()
.cfdPath(path.toString())
.ps(ps)
.setAnalyzer(
new WekaAnalyzer(Class.forName("weka.classifiers.functions.SMO").newInstance()))
.numThreads(4)
.analysisType(FullAPI.analysisType.TRAIN_TEST_UNKNOWN)
.useCache(false)
.build();
long bef1 = System.currentTimeMillis();
test.prepareInstances();
test.run();
long aft1 = System.currentTimeMillis();
String results1 = test.getStatString();
System.out.println(results1);
test =
new FullAPI.Builder()
.cfdPath(path.toString())
.ps(ps)
.setAnalyzer(
new WekaAnalyzer(Class.forName("weka.classifiers.functions.SMO").newInstance()))
.numThreads(4)
.analysisType(FullAPI.analysisType.TRAIN_TEST_UNKNOWN)
.useCache(false)
.build();
long bef2 = System.currentTimeMillis();
test.prepareInstances();
test.run();
long aft2 = System.currentTimeMillis();
String results2 = test.getStatString();
System.out.println(results2);
long time1 = aft1 - bef1;
long time2 = aft2 - bef2;
// This assertion may be too lenient. Just trying to make sure that the time it takes the
// second time is close to the time it takes the first time (in other words, so we know for sure
// it did not use any extracted features)
double percentDiff = (double) (Math.abs(time1 - time2)) / time2;
System.out.println("Percent difference between two runs: " + percentDiff);
File cache = Paths.get(JSANConstants.JSAN_CACHE).toFile();
File[] contents = cache.listFiles();
Assert.assertNotNull("No files in cache. Chunking directory should be there.", contents);
boolean foundChunkingDirectory = false;
for (File f : contents) {
if (f.isDirectory() && f.getName().equals(CacheTests.CHUNKED_DIR_NAME))
foundChunkingDirectory = true;
else {
Assert.fail("The caching system made directories/files when caching was turned off.");
}
}
Assert.assertTrue(
"Chunking was turned on, but no chunking directory was found.", foundChunkingDirectory);
}
/**
* Test to make sure that chunking does not change the results of the cache. It's a long test, but
* it's very important.
*
* @throws ClassNotFoundException
* @throws IllegalAccessException
* @throws InstantiationException
*/
@Test
public void testWithChunking()
throws InstantiationException, IllegalAccessException, ClassNotFoundException {
// the max difference allowed between chunked / non-chunked.
double maxDifferential = 0.05;
ProblemSet ps =
new ProblemSet(
Paths.get(JSANConstants.JSAN_PROBLEMSETS_PREFIX, "drexel_1_small.xml").toString());
Document d = ps.trainDocAt("a", "a_01.txt");
Assert.assertNotNull("No document a_01.txt found.", d);
ps.removeTrainDocAt("a", d);
ps.addTestDoc("a", d);
Path path =
Paths.get(JSANConstants.JSAN_FEATURESETS_PREFIX, "writeprints_feature_set_limited.xml");
FullAPI test =
new FullAPI.Builder()
.cfdPath(path.toString())
.ps(ps)
.setAnalyzer(
new WekaAnalyzer(Class.forName("weka.classifiers.functions.SMO").newInstance()))
.numThreads(4)
.analysisType(FullAPI.analysisType.TRAIN_TEST_UNKNOWN)
.build();
test.prepareInstances();
test.run();
String results1 = test.getStatString();
System.out.println(results1);
test =
new FullAPI.Builder()
.cfdPath(path.toString())
.ps(ps)
.setAnalyzer(
new WekaAnalyzer(Class.forName("weka.classifiers.functions.SMO").newInstance()))
.numThreads(4)
.analysisType(FullAPI.analysisType.TRAIN_TEST_UNKNOWN)
.build();
test.prepareInstances();
test.run();
String results2 = test.getStatString();
System.out.println(results2);
Assert.assertEquals("Cached results different from non-cached results", results1, results2);
// make it rechunk, then test to make sure the results are the same.
deleteRecursive(Paths.get(JSANConstants.JSAN_CHUNK_DIR).toFile(), true);
test =
new FullAPI.Builder()
.cfdPath(path.toString())
.ps(ps)
.setAnalyzer(
new WekaAnalyzer(Class.forName("weka.classifiers.functions.SMO").newInstance()))
.numThreads(4)
.analysisType(FullAPI.analysisType.TRAIN_TEST_UNKNOWN)
.build();
test.prepareInstances();
test.run();
String results3 = test.getStatString();
System.out.println(results3);
Assert.assertEquals("Cached results different from non-cached results", results2, results3);
// now, keep the chunks, but delete the cache, and try again.
File cacheDir = Paths.get(JSANConstants.JSAN_CACHE).toFile();
for (File f : cacheDir.listFiles()) {
if (f.getName() != "chunked") {
deleteRecursive(f, true);
}
}
test =
new FullAPI.Builder()
.cfdPath(path.toString())
.ps(ps)
.setAnalyzer(
new WekaAnalyzer(Class.forName("weka.classifiers.functions.SMO").newInstance()))
.numThreads(4)
.analysisType(FullAPI.analysisType.TRAIN_TEST_UNKNOWN)
.build();
test.prepareInstances();
test.run();
String results4 = test.getStatString();
System.out.println(results4);
Assert.assertEquals("Cached results different from non-cached results", results3, results4);
// ----------------------------------------------------------------------------
// Turn off chunking and make sure the results are the same
// There is good reason to test to make sure the results for this test are
// (close to) the same, regardless of whether chunking is on or not.
// ----------------------------------------------------------------------------
deleteRecursive(Paths.get(JSANConstants.JSAN_CACHE).toFile(), true);
Chunker.shouldChunkTrainDocs(false);
ps =
new ProblemSet(
Paths.get(JSANConstants.JSAN_PROBLEMSETS_PREFIX, "drexel_1_small.xml").toString());
d = ps.trainDocAt("a", "a_01.txt");
Assert.assertNotNull("No document a_01.txt found.", d);
ps.removeTrainDocAt("a", d);
ps.addTestDoc("a", d);
path = Paths.get(JSANConstants.JSAN_FEATURESETS_PREFIX, "writeprints_feature_set_limited.xml");
test =
new FullAPI.Builder()
.cfdPath(path.toString())
.ps(ps)
.setAnalyzer(
new WekaAnalyzer(Class.forName("weka.classifiers.functions.SMO").newInstance()))
.numThreads(4)
.analysisType(FullAPI.analysisType.TRAIN_TEST_UNKNOWN)
.build();
test.prepareInstances();
test.run();
results4 = results4.trim();
String res4 = results4.substring(results4.lastIndexOf("\n") + 1);
String[] allRes4 = Pattern.compile("|", Pattern.LITERAL).split(res4);
String results5 = test.getStatString();
System.out.println(results5);
String res5 = results5.trim().substring(results5.trim().lastIndexOf("\n") + 1);
String[] allRes5 = Pattern.compile("|", Pattern.LITERAL).split(res5);
for (int i = 1; i < allRes4.length; i++) {
double num = Double.parseDouble(allRes4[i].trim().replace(" +", ""));
double num2 = Double.parseDouble(allRes5[i].trim().replace(" +", ""));
Assert.assertTrue(
"There was a large difference between results for chunked / non-chunked."
+ " \nChunked: "
+ num
+ "; Non-chunked: "
+ num2,
(Math.abs((num - num2)) / num2 <= maxDifferential));
}
// Changed test to the one above.. since there WILL be differences between chunked/non-chunked,
// we should just make sure that the difference isn't huge.
// Assert.assertEquals("Chunked results different from non-chunked results", results4,
// results5);
// now do it again with the cache
test =
new FullAPI.Builder()
.cfdPath(path.toString())
.ps(ps)
.setAnalyzer(
new WekaAnalyzer(Class.forName("weka.classifiers.functions.SMO").newInstance()))
.numThreads(4)
.analysisType(FullAPI.analysisType.TRAIN_TEST_UNKNOWN)
.build();
test.prepareInstances();
test.run();
String results6 = test.getStatString();
System.out.println(results6);
Assert.assertEquals("Cached results different from non-cached results", results5, results6);
// make it rechunk, then test to make sure the results are the same.
deleteRecursive(Paths.get(JSANConstants.JSAN_CHUNK_DIR).toFile(), true);
test =
new FullAPI.Builder()
.cfdPath(path.toString())
.ps(ps)
.setAnalyzer(
new WekaAnalyzer(Class.forName("weka.classifiers.functions.SMO").newInstance()))
.numThreads(4)
.analysisType(FullAPI.analysisType.TRAIN_TEST_UNKNOWN)
.build();
test.prepareInstances();
test.run();
String results7 = test.getStatString();
System.out.println(results7);
Assert.assertEquals("Cached results different from non-cached results", results6, results7);
// now, keep the chunks, but delete the cache, and try again.
for (File f : cacheDir.listFiles()) {
if (f.getName() != "chunked") {
deleteRecursive(f, true);
}
}
test =
new FullAPI.Builder()
.cfdPath(path.toString())
.ps(ps)
.setAnalyzer(
new WekaAnalyzer(Class.forName("weka.classifiers.functions.SMO").newInstance()))
.numThreads(4)
.analysisType(FullAPI.analysisType.TRAIN_TEST_UNKNOWN)
.build();
test.prepareInstances();
test.run();
String results8 = test.getStatString();
System.out.println(results8);
Assert.assertEquals("Cached results different from non-cached results", results7, results8);
}