/**
* 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);
}
/** Before each test, do some preparation. */
@Before
public void setUp() {
Chunker.shouldChunkTrainDocs(true);
deleteRecursive(Paths.get(JSANConstants.JSAN_CACHE).toFile(), true);
deleteRecursive(Paths.get(JSANConstants.JUNIT_RESOURCE_PACKAGE, "temp").toFile(), false);
}
/** Stuff to do after we're all done. */
@AfterClass
public static void testTearDown() {
Chunker.shouldChunkTrainDocs(true);
deleteRecursive(Paths.get(JSANConstants.JSAN_CACHE).toFile(), true);
deleteRecursive(Paths.get(JSANConstants.JUNIT_RESOURCE_PACKAGE, "temp").toFile(), false);
}
/**
* 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);
}