| 1 | /* |
|---|
| 2 | * This program is free software; you can redistribute it and/or modify |
|---|
| 3 | * it under the terms of the GNU General Public License as published by |
|---|
| 4 | * the Free Software Foundation; either version 2 of the License, or |
|---|
| 5 | * (at your option) any later version. |
|---|
| 6 | * |
|---|
| 7 | * This program is distributed in the hope that it will be useful, |
|---|
| 8 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
|---|
| 9 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
|---|
| 10 | * GNU General Public License for more details. |
|---|
| 11 | * |
|---|
| 12 | * You should have received a copy of the GNU General Public License |
|---|
| 13 | * along with this program; if not, write to the Free Software |
|---|
| 14 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
|---|
| 15 | */ |
|---|
| 16 | |
|---|
| 17 | /* |
|---|
| 18 | * Copyright (C) 2006 University of Waikato |
|---|
| 19 | */ |
|---|
| 20 | |
|---|
| 21 | package weka.classifiers.functions.supportVector; |
|---|
| 22 | |
|---|
| 23 | import weka.core.Attribute; |
|---|
| 24 | import weka.core.CheckGOE; |
|---|
| 25 | import weka.core.CheckOptionHandler; |
|---|
| 26 | import weka.core.Instances; |
|---|
| 27 | import weka.core.OptionHandler; |
|---|
| 28 | import weka.core.CheckScheme.PostProcessor; |
|---|
| 29 | import weka.test.Regression; |
|---|
| 30 | |
|---|
| 31 | import junit.framework.TestCase; |
|---|
| 32 | |
|---|
| 33 | /** |
|---|
| 34 | * Abstract Test class for Kernels. Internally it uses the class |
|---|
| 35 | * <code>CheckKernel</code> to determine success or failure of the |
|---|
| 36 | * tests. It follows basically the <code>testsPerClassType</code> method. |
|---|
| 37 | * |
|---|
| 38 | * @author FracPete (fracpete at waikato dot ac dot nz) |
|---|
| 39 | * @version $Revision: 1.3 $ |
|---|
| 40 | * |
|---|
| 41 | * @see CheckKernel |
|---|
| 42 | * @see CheckKernel#testsPerClassType(int, boolean, boolean) |
|---|
| 43 | * @see PostProcessor |
|---|
| 44 | */ |
|---|
| 45 | public abstract class AbstractKernelTest |
|---|
| 46 | extends TestCase { |
|---|
| 47 | |
|---|
| 48 | /** The Kernel to be tested */ |
|---|
| 49 | protected Kernel m_Kernel; |
|---|
| 50 | |
|---|
| 51 | /** For testing the Kernel */ |
|---|
| 52 | protected CheckKernel m_Tester; |
|---|
| 53 | |
|---|
| 54 | /** whether Kernel handles weighted instances */ |
|---|
| 55 | protected boolean m_weightedInstancesHandler; |
|---|
| 56 | |
|---|
| 57 | /** whether Kernel handles multi-instance data */ |
|---|
| 58 | protected boolean m_multiInstanceHandler; |
|---|
| 59 | |
|---|
| 60 | /** the number of classes to test with testNClasses() |
|---|
| 61 | * @see #testNClasses() */ |
|---|
| 62 | protected int m_NClasses; |
|---|
| 63 | |
|---|
| 64 | /** whether to run CheckKernel in DEBUG mode */ |
|---|
| 65 | protected boolean DEBUG = false; |
|---|
| 66 | |
|---|
| 67 | /** the attribute type with the lowest value */ |
|---|
| 68 | protected final static int FIRST_CLASSTYPE = Attribute.NUMERIC; |
|---|
| 69 | |
|---|
| 70 | /** the attribute type with the highest value */ |
|---|
| 71 | protected final static int LAST_CLASSTYPE = Attribute.RELATIONAL; |
|---|
| 72 | |
|---|
| 73 | /** wether Kernel can predict nominal attributes (array index is attribute type of class) */ |
|---|
| 74 | protected boolean[] m_NominalPredictors; |
|---|
| 75 | |
|---|
| 76 | /** wether Kernel can predict numeric attributes (array index is attribute type of class) */ |
|---|
| 77 | protected boolean[] m_NumericPredictors; |
|---|
| 78 | |
|---|
| 79 | /** wether Kernel can predict string attributes (array index is attribute type of class) */ |
|---|
| 80 | protected boolean[] m_StringPredictors; |
|---|
| 81 | |
|---|
| 82 | /** wether Kernel can predict date attributes (array index is attribute type of class) */ |
|---|
| 83 | protected boolean[] m_DatePredictors; |
|---|
| 84 | |
|---|
| 85 | /** wether Kernel can predict relational attributes (array index is attribute type of class) */ |
|---|
| 86 | protected boolean[] m_RelationalPredictors; |
|---|
| 87 | |
|---|
| 88 | /** whether Kernel handles missing values */ |
|---|
| 89 | protected boolean[] m_handleMissingPredictors; |
|---|
| 90 | |
|---|
| 91 | /** whether Kernel handles class with only missing values */ |
|---|
| 92 | protected boolean[] m_handleMissingClass; |
|---|
| 93 | |
|---|
| 94 | /** whether Kernel handles class as first attribute */ |
|---|
| 95 | protected boolean[] m_handleClassAsFirstAttribute; |
|---|
| 96 | |
|---|
| 97 | /** whether Kernel handles class as second attribute */ |
|---|
| 98 | protected boolean[] m_handleClassAsSecondAttribute; |
|---|
| 99 | |
|---|
| 100 | /** the results of the regression tests */ |
|---|
| 101 | protected String[] m_RegressionResults; |
|---|
| 102 | |
|---|
| 103 | /** the OptionHandler tester */ |
|---|
| 104 | protected CheckOptionHandler m_OptionTester; |
|---|
| 105 | |
|---|
| 106 | /** for testing GOE stuff */ |
|---|
| 107 | protected CheckGOE m_GOETester; |
|---|
| 108 | |
|---|
| 109 | /** |
|---|
| 110 | * Constructs the <code>AbstractKernelTest</code>. Called by subclasses. |
|---|
| 111 | * |
|---|
| 112 | * @param name the name of the test class |
|---|
| 113 | */ |
|---|
| 114 | public AbstractKernelTest(String name) { |
|---|
| 115 | super(name); |
|---|
| 116 | } |
|---|
| 117 | |
|---|
| 118 | /** |
|---|
| 119 | * returns a custom PostProcessor for the CheckKernel datasets, currently |
|---|
| 120 | * only null. |
|---|
| 121 | * |
|---|
| 122 | * @return a custom PostProcessor, if necessary |
|---|
| 123 | * @see PostProcessor |
|---|
| 124 | */ |
|---|
| 125 | protected PostProcessor getPostProcessor() { |
|---|
| 126 | return null; |
|---|
| 127 | } |
|---|
| 128 | |
|---|
| 129 | /** |
|---|
| 130 | * configures the CheckKernel instance used throughout the tests |
|---|
| 131 | * |
|---|
| 132 | * @return the fully configured CheckKernel instance used for testing |
|---|
| 133 | */ |
|---|
| 134 | protected CheckKernel getTester() { |
|---|
| 135 | CheckKernel result; |
|---|
| 136 | |
|---|
| 137 | result = new CheckKernel(); |
|---|
| 138 | result.setSilent(true); |
|---|
| 139 | result.setKernel(m_Kernel); |
|---|
| 140 | result.setNumInstances(20); |
|---|
| 141 | result.setDebug(DEBUG); |
|---|
| 142 | result.setPostProcessor(getPostProcessor()); |
|---|
| 143 | |
|---|
| 144 | return result; |
|---|
| 145 | } |
|---|
| 146 | |
|---|
| 147 | /** |
|---|
| 148 | * Configures the CheckOptionHandler uses for testing the optionhandling. |
|---|
| 149 | * Sets the Kernel return from the getKernel() method. |
|---|
| 150 | * |
|---|
| 151 | * @return the fully configured CheckOptionHandler |
|---|
| 152 | * @see #getKernel() |
|---|
| 153 | */ |
|---|
| 154 | protected CheckOptionHandler getOptionTester() { |
|---|
| 155 | CheckOptionHandler result; |
|---|
| 156 | |
|---|
| 157 | result = new CheckOptionHandler(); |
|---|
| 158 | if (getKernel() instanceof OptionHandler) |
|---|
| 159 | result.setOptionHandler((OptionHandler) getKernel()); |
|---|
| 160 | else |
|---|
| 161 | result.setOptionHandler(null); |
|---|
| 162 | result.setUserOptions(new String[0]); |
|---|
| 163 | result.setSilent(true); |
|---|
| 164 | |
|---|
| 165 | return result; |
|---|
| 166 | } |
|---|
| 167 | |
|---|
| 168 | /** |
|---|
| 169 | * Configures the CheckGOE used for testing GOE stuff. |
|---|
| 170 | * Sets the Kernel returned from the getKernel() method. |
|---|
| 171 | * |
|---|
| 172 | * @return the fully configured CheckGOE |
|---|
| 173 | * @see #getKernel() |
|---|
| 174 | */ |
|---|
| 175 | protected CheckGOE getGOETester() { |
|---|
| 176 | CheckGOE result; |
|---|
| 177 | |
|---|
| 178 | result = new CheckGOE(); |
|---|
| 179 | result.setObject(getKernel()); |
|---|
| 180 | result.setSilent(true); |
|---|
| 181 | |
|---|
| 182 | return result; |
|---|
| 183 | } |
|---|
| 184 | |
|---|
| 185 | /** |
|---|
| 186 | * Called by JUnit before each test method. This implementation creates |
|---|
| 187 | * the default Kernel to test and loads a test set of Instances. |
|---|
| 188 | * |
|---|
| 189 | * @exception Exception if an error occurs reading the example instances. |
|---|
| 190 | */ |
|---|
| 191 | protected void setUp() throws Exception { |
|---|
| 192 | m_Kernel = getKernel(); |
|---|
| 193 | m_Tester = getTester(); |
|---|
| 194 | m_OptionTester = getOptionTester(); |
|---|
| 195 | m_GOETester = getGOETester(); |
|---|
| 196 | |
|---|
| 197 | m_weightedInstancesHandler = m_Tester.weightedInstancesHandler()[0]; |
|---|
| 198 | m_multiInstanceHandler = m_Tester.multiInstanceHandler()[0]; |
|---|
| 199 | m_NominalPredictors = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 200 | m_NumericPredictors = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 201 | m_StringPredictors = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 202 | m_DatePredictors = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 203 | m_RelationalPredictors = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 204 | m_handleMissingPredictors = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 205 | m_handleMissingClass = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 206 | m_handleClassAsFirstAttribute = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 207 | m_handleClassAsSecondAttribute = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 208 | m_RegressionResults = new String[LAST_CLASSTYPE + 1]; |
|---|
| 209 | m_NClasses = 4; |
|---|
| 210 | |
|---|
| 211 | // initialize attributes |
|---|
| 212 | checkAttributes(true, false, false, false, false, false); |
|---|
| 213 | checkAttributes(false, true, false, false, false, false); |
|---|
| 214 | checkAttributes(false, false, true, false, false, false); |
|---|
| 215 | checkAttributes(false, false, false, true, false, false); |
|---|
| 216 | checkAttributes(false, false, false, false, true, false); |
|---|
| 217 | |
|---|
| 218 | // initialize missing values handling |
|---|
| 219 | for (int i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
|---|
| 220 | // does the scheme support this type of class at all? |
|---|
| 221 | if (!canPredict(i)) |
|---|
| 222 | continue; |
|---|
| 223 | |
|---|
| 224 | // 20% missing |
|---|
| 225 | m_handleMissingPredictors[i] = checkMissingPredictors(i, 20, false); |
|---|
| 226 | m_handleMissingClass[i] = checkMissingClass(i, 20, false); |
|---|
| 227 | } |
|---|
| 228 | } |
|---|
| 229 | |
|---|
| 230 | /** Called by JUnit after each test method */ |
|---|
| 231 | protected void tearDown() { |
|---|
| 232 | m_Kernel = null; |
|---|
| 233 | m_Tester = null; |
|---|
| 234 | m_OptionTester = null; |
|---|
| 235 | m_GOETester = null; |
|---|
| 236 | |
|---|
| 237 | m_weightedInstancesHandler = false; |
|---|
| 238 | m_NominalPredictors = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 239 | m_NumericPredictors = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 240 | m_StringPredictors = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 241 | m_DatePredictors = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 242 | m_RelationalPredictors = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 243 | m_handleMissingPredictors = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 244 | m_handleMissingClass = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 245 | m_handleClassAsFirstAttribute = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 246 | m_handleClassAsSecondAttribute = new boolean[LAST_CLASSTYPE + 1]; |
|---|
| 247 | m_RegressionResults = new String[LAST_CLASSTYPE + 1]; |
|---|
| 248 | m_NClasses = 4; |
|---|
| 249 | } |
|---|
| 250 | |
|---|
| 251 | /** |
|---|
| 252 | * Used to create an instance of a specific Kernel. |
|---|
| 253 | * |
|---|
| 254 | * @return a suitably configured <code>Kernel</code> value |
|---|
| 255 | */ |
|---|
| 256 | public abstract Kernel getKernel(); |
|---|
| 257 | |
|---|
| 258 | /** |
|---|
| 259 | * checks whether at least one attribute type can be handled with the |
|---|
| 260 | * given class type |
|---|
| 261 | * |
|---|
| 262 | * @param type the class type to check for |
|---|
| 263 | * @return true if at least one attribute type can be predicted with |
|---|
| 264 | * the given class |
|---|
| 265 | */ |
|---|
| 266 | protected boolean canPredict(int type) { |
|---|
| 267 | return m_NominalPredictors[type] |
|---|
| 268 | || m_NumericPredictors[type] |
|---|
| 269 | || m_StringPredictors[type] |
|---|
| 270 | || m_DatePredictors[type] |
|---|
| 271 | || m_RelationalPredictors[type]; |
|---|
| 272 | } |
|---|
| 273 | |
|---|
| 274 | /** |
|---|
| 275 | * returns a string for the class type |
|---|
| 276 | * |
|---|
| 277 | * @param type the class type |
|---|
| 278 | * @return the class type as string |
|---|
| 279 | */ |
|---|
| 280 | protected String getClassTypeString(int type) { |
|---|
| 281 | return CheckKernel.attributeTypeToString(type); |
|---|
| 282 | } |
|---|
| 283 | |
|---|
| 284 | /** |
|---|
| 285 | * tests whether the Kernel can handle certain attributes and if not, |
|---|
| 286 | * if the exception is OK |
|---|
| 287 | * |
|---|
| 288 | * @param nom to check for nominal attributes |
|---|
| 289 | * @param num to check for numeric attributes |
|---|
| 290 | * @param str to check for string attributes |
|---|
| 291 | * @param dat to check for date attributes |
|---|
| 292 | * @param rel to check for relational attributes |
|---|
| 293 | * @param allowFail whether a junit fail can be executed |
|---|
| 294 | * @see CheckKernel#canPredict(boolean, boolean, boolean, boolean, boolean, boolean, int) |
|---|
| 295 | * @see CheckKernel#testsPerClassType(int, boolean, boolean) |
|---|
| 296 | */ |
|---|
| 297 | protected void checkAttributes(boolean nom, boolean num, boolean str, |
|---|
| 298 | boolean dat, boolean rel, |
|---|
| 299 | boolean allowFail) { |
|---|
| 300 | boolean[] result; |
|---|
| 301 | String att; |
|---|
| 302 | int i; |
|---|
| 303 | |
|---|
| 304 | // determine text for type of attributes |
|---|
| 305 | att = ""; |
|---|
| 306 | if (nom) |
|---|
| 307 | att = "nominal"; |
|---|
| 308 | else if (num) |
|---|
| 309 | att = "numeric"; |
|---|
| 310 | else if (str) |
|---|
| 311 | att = "string"; |
|---|
| 312 | else if (dat) |
|---|
| 313 | att = "date"; |
|---|
| 314 | else if (rel) |
|---|
| 315 | att = "relational"; |
|---|
| 316 | |
|---|
| 317 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
|---|
| 318 | result = m_Tester.canPredict(nom, num, str, dat, rel, m_multiInstanceHandler, i); |
|---|
| 319 | if (nom) |
|---|
| 320 | m_NominalPredictors[i] = result[0]; |
|---|
| 321 | else if (num) |
|---|
| 322 | m_NumericPredictors[i] = result[0]; |
|---|
| 323 | else if (str) |
|---|
| 324 | m_StringPredictors[i] = result[0]; |
|---|
| 325 | else if (dat) |
|---|
| 326 | m_DatePredictors[i] = result[0]; |
|---|
| 327 | else if (rel) |
|---|
| 328 | m_RelationalPredictors[i] = result[0]; |
|---|
| 329 | |
|---|
| 330 | if (!result[0] && !result[1] && allowFail) |
|---|
| 331 | fail("Error handling " + att + " attributes (" + getClassTypeString(i) |
|---|
| 332 | + " class)!"); |
|---|
| 333 | } |
|---|
| 334 | } |
|---|
| 335 | |
|---|
| 336 | /** |
|---|
| 337 | * tests whether the Kernel can handle different types of attributes and |
|---|
| 338 | * if not, if the exception is OK |
|---|
| 339 | * |
|---|
| 340 | * @see #checkAttributes(boolean, boolean, boolean, boolean, boolean, boolean) |
|---|
| 341 | */ |
|---|
| 342 | public void testAttributes() { |
|---|
| 343 | // nominal |
|---|
| 344 | checkAttributes(true, false, false, false, false, true); |
|---|
| 345 | // numeric |
|---|
| 346 | checkAttributes(false, true, false, false, false, true); |
|---|
| 347 | // string |
|---|
| 348 | checkAttributes(false, false, true, false, false, true); |
|---|
| 349 | // date |
|---|
| 350 | checkAttributes(false, false, false, true, false, true); |
|---|
| 351 | // relational |
|---|
| 352 | if (!m_multiInstanceHandler) |
|---|
| 353 | checkAttributes(false, false, false, false, true, true); |
|---|
| 354 | } |
|---|
| 355 | |
|---|
| 356 | /** |
|---|
| 357 | * tests whether the scheme declares a serialVersionUID. |
|---|
| 358 | */ |
|---|
| 359 | public void testSerialVersionUID() { |
|---|
| 360 | boolean[] result; |
|---|
| 361 | |
|---|
| 362 | result = m_Tester.declaresSerialVersionUID(); |
|---|
| 363 | |
|---|
| 364 | if (!result[0]) |
|---|
| 365 | fail("Doesn't declare serialVersionUID!"); |
|---|
| 366 | } |
|---|
| 367 | |
|---|
| 368 | /** |
|---|
| 369 | * tests whether the Kernel handles instance weights correctly |
|---|
| 370 | * |
|---|
| 371 | * @see CheckKernel#instanceWeights(boolean, boolean, boolean, boolean, boolean, boolean, int) |
|---|
| 372 | * @see CheckKernel#testsPerClassType(int, boolean, boolean) |
|---|
| 373 | */ |
|---|
| 374 | public void testInstanceWeights() { |
|---|
| 375 | boolean[] result; |
|---|
| 376 | int i; |
|---|
| 377 | |
|---|
| 378 | if (m_weightedInstancesHandler) { |
|---|
| 379 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
|---|
| 380 | // does the Kernel support this type of class at all? |
|---|
| 381 | if (!canPredict(i)) |
|---|
| 382 | continue; |
|---|
| 383 | |
|---|
| 384 | result = m_Tester.instanceWeights( |
|---|
| 385 | m_NominalPredictors[i], |
|---|
| 386 | m_NumericPredictors[i], |
|---|
| 387 | m_StringPredictors[i], |
|---|
| 388 | m_DatePredictors[i], |
|---|
| 389 | m_RelationalPredictors[i], |
|---|
| 390 | m_multiInstanceHandler, |
|---|
| 391 | i); |
|---|
| 392 | |
|---|
| 393 | if (!result[0]) |
|---|
| 394 | System.err.println("Error handling instance weights (" + getClassTypeString(i) |
|---|
| 395 | + " class)!"); |
|---|
| 396 | } |
|---|
| 397 | } |
|---|
| 398 | } |
|---|
| 399 | |
|---|
| 400 | /** |
|---|
| 401 | * tests whether Kernel handles N classes |
|---|
| 402 | * |
|---|
| 403 | * @see CheckKernel#canHandleNClasses(boolean, boolean, boolean, boolean, boolean, boolean, int) |
|---|
| 404 | * @see CheckKernel#testsPerClassType(int, boolean, boolean) |
|---|
| 405 | * @see #m_NClasses |
|---|
| 406 | */ |
|---|
| 407 | public void testNClasses() { |
|---|
| 408 | boolean[] result; |
|---|
| 409 | |
|---|
| 410 | if (!canPredict(Attribute.NOMINAL)) |
|---|
| 411 | return; |
|---|
| 412 | |
|---|
| 413 | result = m_Tester.canHandleNClasses( |
|---|
| 414 | m_NominalPredictors[Attribute.NOMINAL], |
|---|
| 415 | m_NumericPredictors[Attribute.NOMINAL], |
|---|
| 416 | m_StringPredictors[Attribute.NOMINAL], |
|---|
| 417 | m_DatePredictors[Attribute.NOMINAL], |
|---|
| 418 | m_RelationalPredictors[Attribute.NOMINAL], |
|---|
| 419 | m_multiInstanceHandler, |
|---|
| 420 | m_NClasses); |
|---|
| 421 | |
|---|
| 422 | if (!result[0] && !result[1]) |
|---|
| 423 | fail("Error handling " + m_NClasses + " classes!"); |
|---|
| 424 | } |
|---|
| 425 | |
|---|
| 426 | /** |
|---|
| 427 | * checks whether the Kernel can handle the class attribute at a given |
|---|
| 428 | * position (0-based index, -1 means last). |
|---|
| 429 | * |
|---|
| 430 | * @param type the class type |
|---|
| 431 | * @param position the position of the class attribute (0-based, -1 means last) |
|---|
| 432 | * @return true if the Kernel can handle it |
|---|
| 433 | */ |
|---|
| 434 | protected boolean checkClassAsNthAttribute(int type, int position) { |
|---|
| 435 | boolean[] result; |
|---|
| 436 | String indexStr; |
|---|
| 437 | |
|---|
| 438 | result = m_Tester.canHandleClassAsNthAttribute( |
|---|
| 439 | m_NominalPredictors[type], |
|---|
| 440 | m_NumericPredictors[type], |
|---|
| 441 | m_StringPredictors[type], |
|---|
| 442 | m_DatePredictors[type], |
|---|
| 443 | m_RelationalPredictors[type], |
|---|
| 444 | m_multiInstanceHandler, |
|---|
| 445 | type, |
|---|
| 446 | position); |
|---|
| 447 | |
|---|
| 448 | if (position == -1) |
|---|
| 449 | indexStr = "last"; |
|---|
| 450 | else |
|---|
| 451 | indexStr = (position + 1) + "."; |
|---|
| 452 | |
|---|
| 453 | if (!result[0] && !result[1]) |
|---|
| 454 | fail("Error handling class as " + indexStr + " attribute (" |
|---|
| 455 | + getClassTypeString(type) + " class)!"); |
|---|
| 456 | |
|---|
| 457 | return result[0]; |
|---|
| 458 | } |
|---|
| 459 | |
|---|
| 460 | /** |
|---|
| 461 | * Tests whether the Kernel can handle class attributes as Nth |
|---|
| 462 | * attribute. In case of multi-instance Kernels it performs no tests, |
|---|
| 463 | * since the multi-instance data has a fixed format (bagID,bag,class). |
|---|
| 464 | * |
|---|
| 465 | * @see CheckKernel#canHandleClassAsNthAttribute(boolean, boolean, boolean, boolean, boolean, boolean, int, int) |
|---|
| 466 | * @see CheckKernel#testsPerClassType(int, boolean, boolean) |
|---|
| 467 | */ |
|---|
| 468 | public void testClassAsNthAttribute() { |
|---|
| 469 | int i; |
|---|
| 470 | |
|---|
| 471 | // multi-Instance data has fixed format! |
|---|
| 472 | if (m_multiInstanceHandler) |
|---|
| 473 | return; |
|---|
| 474 | |
|---|
| 475 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
|---|
| 476 | // does the Kernel support this type of class at all? |
|---|
| 477 | if (!canPredict(i)) |
|---|
| 478 | continue; |
|---|
| 479 | |
|---|
| 480 | // first attribute |
|---|
| 481 | m_handleClassAsFirstAttribute[i] = checkClassAsNthAttribute(i, 0); |
|---|
| 482 | |
|---|
| 483 | // second attribute |
|---|
| 484 | m_handleClassAsSecondAttribute[i] = checkClassAsNthAttribute(i, 1); |
|---|
| 485 | } |
|---|
| 486 | } |
|---|
| 487 | |
|---|
| 488 | /** |
|---|
| 489 | * tests whether the Kernel can handle zero training instances |
|---|
| 490 | * |
|---|
| 491 | * @see CheckKernel#canHandleZeroTraining(boolean, boolean, boolean, boolean, boolean, boolean, int) |
|---|
| 492 | * @see CheckKernel#testsPerClassType(int, boolean, boolean) |
|---|
| 493 | */ |
|---|
| 494 | public void testZeroTraining() { |
|---|
| 495 | boolean[] result; |
|---|
| 496 | int i; |
|---|
| 497 | |
|---|
| 498 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
|---|
| 499 | // does the Kernel support this type of class at all? |
|---|
| 500 | if (!canPredict(i)) |
|---|
| 501 | continue; |
|---|
| 502 | |
|---|
| 503 | result = m_Tester.canHandleZeroTraining( |
|---|
| 504 | m_NominalPredictors[i], |
|---|
| 505 | m_NumericPredictors[i], |
|---|
| 506 | m_StringPredictors[i], |
|---|
| 507 | m_DatePredictors[i], |
|---|
| 508 | m_RelationalPredictors[i], |
|---|
| 509 | m_multiInstanceHandler, |
|---|
| 510 | i); |
|---|
| 511 | |
|---|
| 512 | if (!result[0] && !result[1]) |
|---|
| 513 | fail("Error handling zero training instances (" + getClassTypeString(i) |
|---|
| 514 | + " class)!"); |
|---|
| 515 | } |
|---|
| 516 | } |
|---|
| 517 | |
|---|
| 518 | /** |
|---|
| 519 | * checks whether the Kernel can handle the given percentage of |
|---|
| 520 | * missing predictors |
|---|
| 521 | * |
|---|
| 522 | * @param type the class type |
|---|
| 523 | * @param percent the percentage of missing predictors |
|---|
| 524 | * @param allowFail if true a fail statement may be executed |
|---|
| 525 | * @return true if the Kernel can handle it |
|---|
| 526 | */ |
|---|
| 527 | protected boolean checkMissingPredictors(int type, int percent, boolean allowFail) { |
|---|
| 528 | boolean[] result; |
|---|
| 529 | |
|---|
| 530 | result = m_Tester.canHandleMissing( |
|---|
| 531 | m_NominalPredictors[type], |
|---|
| 532 | m_NumericPredictors[type], |
|---|
| 533 | m_StringPredictors[type], |
|---|
| 534 | m_DatePredictors[type], |
|---|
| 535 | m_RelationalPredictors[type], |
|---|
| 536 | m_multiInstanceHandler, |
|---|
| 537 | type, |
|---|
| 538 | true, |
|---|
| 539 | false, |
|---|
| 540 | percent); |
|---|
| 541 | |
|---|
| 542 | if (allowFail) { |
|---|
| 543 | if (!result[0] && !result[1]) |
|---|
| 544 | fail("Error handling " + percent + "% missing predictors (" |
|---|
| 545 | + getClassTypeString(type) + " class)!"); |
|---|
| 546 | } |
|---|
| 547 | |
|---|
| 548 | return result[0]; |
|---|
| 549 | } |
|---|
| 550 | |
|---|
| 551 | /** |
|---|
| 552 | * tests whether the Kernel can handle missing predictors (20% and 100%) |
|---|
| 553 | * |
|---|
| 554 | * @see CheckKernel#canHandleMissing(boolean, boolean, boolean, boolean, boolean, boolean, int, boolean, boolean, int) |
|---|
| 555 | * @see CheckKernel#testsPerClassType(int, boolean, boolean) |
|---|
| 556 | */ |
|---|
| 557 | public void testMissingPredictors() { |
|---|
| 558 | int i; |
|---|
| 559 | |
|---|
| 560 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
|---|
| 561 | // does the Kernel support this type of class at all? |
|---|
| 562 | if (!canPredict(i)) |
|---|
| 563 | continue; |
|---|
| 564 | |
|---|
| 565 | // 20% missing |
|---|
| 566 | checkMissingPredictors(i, 20, true); |
|---|
| 567 | |
|---|
| 568 | // 100% missing |
|---|
| 569 | if (m_handleMissingPredictors[i]) |
|---|
| 570 | checkMissingPredictors(i, 100, true); |
|---|
| 571 | } |
|---|
| 572 | } |
|---|
| 573 | |
|---|
| 574 | /** |
|---|
| 575 | * checks whether the Kernel can handle the given percentage of |
|---|
| 576 | * missing class labels |
|---|
| 577 | * |
|---|
| 578 | * @param type the class type |
|---|
| 579 | * @param percent the percentage of missing class labels |
|---|
| 580 | * @param allowFail if true a fail statement may be executed |
|---|
| 581 | * @return true if the Kernel can handle it |
|---|
| 582 | */ |
|---|
| 583 | protected boolean checkMissingClass(int type, int percent, boolean allowFail) { |
|---|
| 584 | boolean[] result; |
|---|
| 585 | |
|---|
| 586 | result = m_Tester.canHandleMissing( |
|---|
| 587 | m_NominalPredictors[type], |
|---|
| 588 | m_NumericPredictors[type], |
|---|
| 589 | m_StringPredictors[type], |
|---|
| 590 | m_DatePredictors[type], |
|---|
| 591 | m_RelationalPredictors[type], |
|---|
| 592 | m_multiInstanceHandler, |
|---|
| 593 | type, |
|---|
| 594 | false, |
|---|
| 595 | true, |
|---|
| 596 | percent); |
|---|
| 597 | |
|---|
| 598 | if (allowFail) { |
|---|
| 599 | if (!result[0] && !result[1]) |
|---|
| 600 | fail("Error handling " + percent + "% missing class labels (" |
|---|
| 601 | + getClassTypeString(type) + " class)!"); |
|---|
| 602 | } |
|---|
| 603 | |
|---|
| 604 | return result[0]; |
|---|
| 605 | } |
|---|
| 606 | |
|---|
| 607 | /** |
|---|
| 608 | * tests whether the Kernel can handle missing class values (20% and |
|---|
| 609 | * 100%) |
|---|
| 610 | * |
|---|
| 611 | * @see CheckKernel#canHandleMissing(boolean, boolean, boolean, boolean, boolean, boolean, int, boolean, boolean, int) |
|---|
| 612 | * @see CheckKernel#testsPerClassType(int, boolean, boolean) |
|---|
| 613 | */ |
|---|
| 614 | public void testMissingClass() { |
|---|
| 615 | int i; |
|---|
| 616 | |
|---|
| 617 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
|---|
| 618 | // does the Kernel support this type of class at all? |
|---|
| 619 | if (!canPredict(i)) |
|---|
| 620 | continue; |
|---|
| 621 | |
|---|
| 622 | // 20% missing |
|---|
| 623 | checkMissingClass(i, 20, true); |
|---|
| 624 | |
|---|
| 625 | // 100% missing |
|---|
| 626 | if (m_handleMissingClass[i]) |
|---|
| 627 | checkMissingClass(i, 100, true); |
|---|
| 628 | } |
|---|
| 629 | } |
|---|
| 630 | |
|---|
| 631 | /** |
|---|
| 632 | * tests whether the Kernel correctly initializes in the |
|---|
| 633 | * buildKernel method |
|---|
| 634 | * |
|---|
| 635 | * @see CheckKernel#correctBuildInitialisation(boolean, boolean, boolean, boolean, boolean, boolean, int) |
|---|
| 636 | * @see CheckKernel#testsPerClassType(int, boolean, boolean) |
|---|
| 637 | */ |
|---|
| 638 | public void testBuildInitialization() { |
|---|
| 639 | boolean[] result; |
|---|
| 640 | int i; |
|---|
| 641 | |
|---|
| 642 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
|---|
| 643 | // does the Kernel support this type of class at all? |
|---|
| 644 | if (!canPredict(i)) |
|---|
| 645 | continue; |
|---|
| 646 | |
|---|
| 647 | result = m_Tester.correctBuildInitialisation( |
|---|
| 648 | m_NominalPredictors[i], |
|---|
| 649 | m_NumericPredictors[i], |
|---|
| 650 | m_StringPredictors[i], |
|---|
| 651 | m_DatePredictors[i], |
|---|
| 652 | m_RelationalPredictors[i], |
|---|
| 653 | m_multiInstanceHandler, |
|---|
| 654 | i); |
|---|
| 655 | |
|---|
| 656 | if (!result[0] && !result[1]) |
|---|
| 657 | fail("Incorrect build initialization (" + getClassTypeString(i) |
|---|
| 658 | + " class)!"); |
|---|
| 659 | } |
|---|
| 660 | } |
|---|
| 661 | |
|---|
| 662 | /** |
|---|
| 663 | * tests whether the Kernel alters the training set during training. |
|---|
| 664 | * |
|---|
| 665 | * @see CheckKernel#datasetIntegrity(boolean, boolean, boolean, boolean, boolean, boolean, int, boolean, boolean) |
|---|
| 666 | * @see CheckKernel#testsPerClassType(int, boolean, boolean) |
|---|
| 667 | */ |
|---|
| 668 | public void testDatasetIntegrity() { |
|---|
| 669 | boolean[] result; |
|---|
| 670 | int i; |
|---|
| 671 | |
|---|
| 672 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
|---|
| 673 | // does the Kernel support this type of class at all? |
|---|
| 674 | if (!canPredict(i)) |
|---|
| 675 | continue; |
|---|
| 676 | |
|---|
| 677 | result = m_Tester.datasetIntegrity( |
|---|
| 678 | m_NominalPredictors[i], |
|---|
| 679 | m_NumericPredictors[i], |
|---|
| 680 | m_StringPredictors[i], |
|---|
| 681 | m_DatePredictors[i], |
|---|
| 682 | m_RelationalPredictors[i], |
|---|
| 683 | m_multiInstanceHandler, |
|---|
| 684 | i, |
|---|
| 685 | m_handleMissingPredictors[i], |
|---|
| 686 | m_handleMissingClass[i]); |
|---|
| 687 | |
|---|
| 688 | if (!result[0] && !result[1]) |
|---|
| 689 | fail("Training set is altered during training (" |
|---|
| 690 | + getClassTypeString(i) + " class)!"); |
|---|
| 691 | } |
|---|
| 692 | } |
|---|
| 693 | |
|---|
| 694 | /** |
|---|
| 695 | * Builds a model using the current Kernel using the given data and |
|---|
| 696 | * returns the produced output. |
|---|
| 697 | * |
|---|
| 698 | * @param data the instances to test the Kernel on |
|---|
| 699 | * @return a String containing the output of the Kernel. |
|---|
| 700 | */ |
|---|
| 701 | protected String useKernel(Instances data) throws Exception { |
|---|
| 702 | Kernel kernel = null; |
|---|
| 703 | StringBuffer text = new StringBuffer(); |
|---|
| 704 | |
|---|
| 705 | try { |
|---|
| 706 | kernel = Kernel.makeCopy(m_Kernel); |
|---|
| 707 | } |
|---|
| 708 | catch (Exception e) { |
|---|
| 709 | e.printStackTrace(); |
|---|
| 710 | fail("Problem setting up to use Kernel: " + e); |
|---|
| 711 | } |
|---|
| 712 | |
|---|
| 713 | kernel.buildKernel(data); |
|---|
| 714 | for (int n = 0; n < data.numInstances(); n++) { |
|---|
| 715 | for (int i = n; i < data.numInstances(); i++) { |
|---|
| 716 | text.append( |
|---|
| 717 | (n+1) + "-" + (i+1) + ": " |
|---|
| 718 | + kernel.eval(n, i, data.instance(i)) + "\n"); |
|---|
| 719 | } |
|---|
| 720 | } |
|---|
| 721 | |
|---|
| 722 | return text.toString(); |
|---|
| 723 | } |
|---|
| 724 | |
|---|
| 725 | /** |
|---|
| 726 | * Provides a hook for derived classes to further modify the data. Currently, |
|---|
| 727 | * the data is just passed through. |
|---|
| 728 | * |
|---|
| 729 | * @param data the data to process |
|---|
| 730 | * @return the processed data |
|---|
| 731 | */ |
|---|
| 732 | protected Instances process(Instances data) { |
|---|
| 733 | return data; |
|---|
| 734 | } |
|---|
| 735 | |
|---|
| 736 | /** |
|---|
| 737 | * Runs a regression test -- this checks that the output of the tested |
|---|
| 738 | * object matches that in a reference version. When this test is |
|---|
| 739 | * run without any pre-existing reference output, the reference version |
|---|
| 740 | * is created. |
|---|
| 741 | */ |
|---|
| 742 | public void testRegression() throws Exception { |
|---|
| 743 | int i; |
|---|
| 744 | boolean succeeded; |
|---|
| 745 | Regression reg; |
|---|
| 746 | Instances train; |
|---|
| 747 | |
|---|
| 748 | // don't bother if not working correctly |
|---|
| 749 | if (m_Tester.hasClasspathProblems()) |
|---|
| 750 | return; |
|---|
| 751 | |
|---|
| 752 | reg = new Regression(this.getClass()); |
|---|
| 753 | succeeded = false; |
|---|
| 754 | train = null; |
|---|
| 755 | |
|---|
| 756 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
|---|
| 757 | // does the Kernel support this type of class at all? |
|---|
| 758 | if (!canPredict(i)) |
|---|
| 759 | continue; |
|---|
| 760 | |
|---|
| 761 | train = m_Tester.makeTestDataset( |
|---|
| 762 | 42, m_Tester.getNumInstances(), |
|---|
| 763 | m_NominalPredictors[i] ? 2 : 0, |
|---|
| 764 | m_NumericPredictors[i] ? 1 : 0, |
|---|
| 765 | m_StringPredictors[i] ? 1 : 0, |
|---|
| 766 | m_DatePredictors[i] ? 1 : 0, |
|---|
| 767 | m_RelationalPredictors[i] ? 1 : 0, |
|---|
| 768 | 2, |
|---|
| 769 | i, |
|---|
| 770 | m_multiInstanceHandler); |
|---|
| 771 | |
|---|
| 772 | try { |
|---|
| 773 | m_RegressionResults[i] = CheckKernel.attributeTypeToString(i) |
|---|
| 774 | + " class:\n" |
|---|
| 775 | + useKernel(train); |
|---|
| 776 | succeeded = true; |
|---|
| 777 | reg.println(m_RegressionResults[i]); |
|---|
| 778 | } |
|---|
| 779 | catch (Exception e) { |
|---|
| 780 | String msg = e.getMessage().toLowerCase(); |
|---|
| 781 | if (msg.indexOf("not in classpath") > -1) |
|---|
| 782 | return; |
|---|
| 783 | |
|---|
| 784 | m_RegressionResults[i] = null; |
|---|
| 785 | } |
|---|
| 786 | } |
|---|
| 787 | |
|---|
| 788 | if (!succeeded) { |
|---|
| 789 | fail("Problem during regression testing: no successful predictions for any class type"); |
|---|
| 790 | } |
|---|
| 791 | |
|---|
| 792 | try { |
|---|
| 793 | String diff = reg.diff(); |
|---|
| 794 | if (diff == null) { |
|---|
| 795 | System.err.println("Warning: No reference available, creating."); |
|---|
| 796 | } else if (!diff.equals("")) { |
|---|
| 797 | fail("Regression test failed. Difference:\n" + diff); |
|---|
| 798 | } |
|---|
| 799 | } |
|---|
| 800 | catch (java.io.IOException ex) { |
|---|
| 801 | fail("Problem during regression testing.\n" + ex); |
|---|
| 802 | } |
|---|
| 803 | } |
|---|
| 804 | |
|---|
| 805 | /** |
|---|
| 806 | * tests the listing of the options |
|---|
| 807 | */ |
|---|
| 808 | public void testListOptions() { |
|---|
| 809 | if (m_OptionTester.getOptionHandler() != null) { |
|---|
| 810 | if (!m_OptionTester.checkListOptions()) |
|---|
| 811 | fail("Options cannot be listed via listOptions."); |
|---|
| 812 | } |
|---|
| 813 | } |
|---|
| 814 | |
|---|
| 815 | /** |
|---|
| 816 | * tests the setting of the options |
|---|
| 817 | */ |
|---|
| 818 | public void testSetOptions() { |
|---|
| 819 | if (m_OptionTester.getOptionHandler() != null) { |
|---|
| 820 | if (!m_OptionTester.checkSetOptions()) |
|---|
| 821 | fail("setOptions method failed."); |
|---|
| 822 | } |
|---|
| 823 | } |
|---|
| 824 | |
|---|
| 825 | /** |
|---|
| 826 | * tests whether there are any remaining options |
|---|
| 827 | */ |
|---|
| 828 | public void testRemainingOptions() { |
|---|
| 829 | if (m_OptionTester.getOptionHandler() != null) { |
|---|
| 830 | if (!m_OptionTester.checkRemainingOptions()) |
|---|
| 831 | fail("There were 'left-over' options."); |
|---|
| 832 | } |
|---|
| 833 | } |
|---|
| 834 | |
|---|
| 835 | /** |
|---|
| 836 | * tests the whether the user-supplied options stay the same after setting. |
|---|
| 837 | * getting, and re-setting again. |
|---|
| 838 | * |
|---|
| 839 | * @see #getOptionTester() |
|---|
| 840 | */ |
|---|
| 841 | public void testCanonicalUserOptions() { |
|---|
| 842 | if (m_OptionTester.getOptionHandler() != null) { |
|---|
| 843 | if (!m_OptionTester.checkCanonicalUserOptions()) |
|---|
| 844 | fail("setOptions method failed"); |
|---|
| 845 | } |
|---|
| 846 | } |
|---|
| 847 | |
|---|
| 848 | /** |
|---|
| 849 | * tests the resetting of the options to the default ones |
|---|
| 850 | */ |
|---|
| 851 | public void testResettingOptions() { |
|---|
| 852 | if (m_OptionTester.getOptionHandler() != null) { |
|---|
| 853 | if (!m_OptionTester.checkSetOptions()) |
|---|
| 854 | fail("Resetting of options failed"); |
|---|
| 855 | } |
|---|
| 856 | } |
|---|
| 857 | |
|---|
| 858 | /** |
|---|
| 859 | * tests for a globalInfo method |
|---|
| 860 | */ |
|---|
| 861 | public void testGlobalInfo() { |
|---|
| 862 | if (!m_GOETester.checkGlobalInfo()) |
|---|
| 863 | fail("No globalInfo method"); |
|---|
| 864 | } |
|---|
| 865 | |
|---|
| 866 | /** |
|---|
| 867 | * tests the tool tips |
|---|
| 868 | */ |
|---|
| 869 | public void testToolTips() { |
|---|
| 870 | if (!m_GOETester.checkToolTips()) |
|---|
| 871 | fail("Tool tips inconsistent"); |
|---|
| 872 | } |
|---|
| 873 | } |
|---|