/**
* Test to make sure that chunking does not change the results.
*
* @throws ClassNotFoundException
* @throws IllegalAccessException
* @throws InstantiationException
*/
@Test
public void testWithoutChunking()
throws InstantiationException, IllegalAccessException, ClassNotFoundException {
// TODO: this test is basically encapsulated in testWithChunking... so maybe get rid of this?
Chunker.shouldChunkTrainDocs(false);
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);
}
/**
* 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);
}
/*
* Tests to make sure the caching system works.
* Compares the results of using cached/non-cached features.
*
* Takes about 1 minute, depending on the authors selected and how many documents they have.
*/
@Test
public void scalingCached() throws Exception {
// TODO: Tests generally should have no randomness. We should perform these tests on a set of
// static problem sets
// that the test uses every time.
int num_authors = 2;
System.out.println(num_authors);
File source = Paths.get(JSANConstants.JSAN_CORPORA_PREFIX, "drexel_1").toFile();
Assert.assertTrue(
"You don't have the specified corpora available.", source.exists() && source.isDirectory());
// This can be anything. Problem set location
File dest = Paths.get(JSANConstants.JUNIT_RESOURCE_PACKAGE, "temp", "cache_test_.xml").toFile();
GenericProblemSetCreator.createProblemSetsWithSize(source, dest, num_authors, 10);
File[] problem_sets = BekahUtil.listNotHiddenFiles(dest.getParentFile());
for (File problem_set : problem_sets) {
// Delete the cache in the beginning
// File to cache
deleteRecursive(new File(JSANConstants.JSAN_CACHE), false);
try {
Thread.sleep(20);
} catch (InterruptedException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
System.out.println(problem_set);
ProblemSet p = new ProblemSet(problem_set.getAbsolutePath());
Set<String> authors = p.getAuthors();
for (String a : authors) {
System.out.print(a + ",");
}
System.out.println("");
Path path =
Paths.get(JSANConstants.JSAN_FEATURESETS_PREFIX, "writeprints_feature_set_limited.xml");
FullAPI test =
new FullAPI.Builder()
.cfdPath(path.toString())
.psPath(problem_set.getAbsolutePath())
.setAnalyzer(
new WekaAnalyzer(Class.forName("weka.classifiers.functions.SMO").newInstance()))
.numThreads(4)
.analysisType(FullAPI.analysisType.CROSS_VALIDATION)
.build();
long bef1 = System.currentTimeMillis();
test.prepareInstances();
test.run();
ExperimentResults r1 = test.getResults();
long aft1 = System.currentTimeMillis();
String result1 = r1.getStatisticsString();
System.out.println(result1);
FullAPI test2 =
new FullAPI.Builder()
.cfdPath(path.toString())
.psPath(problem_set.getAbsolutePath())
.setAnalyzer(
new WekaAnalyzer(Class.forName("weka.classifiers.functions.SMO").newInstance()))
.numThreads(4)
.analysisType(FullAPI.analysisType.CROSS_VALIDATION)
.build();
long bef2 = System.currentTimeMillis();
test2.prepareInstances();
test2.run();
ExperimentResults r2 = test2.getResults();
long aft2 = System.currentTimeMillis();
String result2 = r2.getStatisticsString();
System.out.println(result2);
Assert.assertEquals("Cached results different from non-cached results", result1, result2);
System.out.print((aft1 - bef1) / 1000 + " s\t");
System.out.println((aft2 - bef2) / 1000 + " s");
}
}
/**
* 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);
}