[4] | 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.attributeSelection; |
---|
| 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.CheckScheme.PostProcessor; |
---|
| 28 | import weka.test.Regression; |
---|
| 29 | |
---|
| 30 | import junit.framework.TestCase; |
---|
| 31 | |
---|
| 32 | /** |
---|
| 33 | * Abstract Test class for attribute selection schemes. Internally it uses the |
---|
| 34 | * class <code>CheckAttributeSelection</code> to determine success or failure |
---|
| 35 | * of the tests. It follows basically the <code>testsPerClassType</code> |
---|
| 36 | * method. |
---|
| 37 | * |
---|
| 38 | * @author FracPete (fracpete at waikato dot ac dot nz) |
---|
| 39 | * @version $Revision: 1.5 $ |
---|
| 40 | * |
---|
| 41 | * @see CheckAttributeSelection |
---|
| 42 | * @see CheckAttributeSelection#testsPerClassType(int, boolean, boolean) |
---|
| 43 | * @see PostProcessor |
---|
| 44 | */ |
---|
| 45 | public abstract class AbstractAttributeSelectionTest |
---|
| 46 | extends TestCase { |
---|
| 47 | |
---|
| 48 | /** The search scheme to be tested */ |
---|
| 49 | protected ASSearch m_Search; |
---|
| 50 | |
---|
| 51 | /** The evaluator to test */ |
---|
| 52 | protected ASEvaluation m_Evaluator; |
---|
| 53 | |
---|
| 54 | /** For testing the attribute selection scheme */ |
---|
| 55 | protected CheckAttributeSelection m_Tester; |
---|
| 56 | |
---|
| 57 | /** whether scheme handles weighted instances */ |
---|
| 58 | protected boolean m_weightedInstancesHandler; |
---|
| 59 | |
---|
| 60 | /** whether scheme handles multi-instance data */ |
---|
| 61 | protected boolean m_multiInstanceHandler; |
---|
| 62 | |
---|
| 63 | /** the number of classes to test with testNClasses() |
---|
| 64 | * @see #testNClasses() */ |
---|
| 65 | protected int m_NClasses; |
---|
| 66 | |
---|
| 67 | /** whether to run CheckAttributeSelection in DEBUG mode */ |
---|
| 68 | protected boolean DEBUG = false; |
---|
| 69 | |
---|
| 70 | /** the attribute type with the lowest value */ |
---|
| 71 | protected final static int FIRST_CLASSTYPE = Attribute.NUMERIC; |
---|
| 72 | |
---|
| 73 | /** the attribute type with the highest value */ |
---|
| 74 | protected final static int LAST_CLASSTYPE = Attribute.RELATIONAL; |
---|
| 75 | |
---|
| 76 | /** wether scheme can predict nominal attributes (array index is attribute type of class) */ |
---|
| 77 | protected boolean[] m_NominalPredictors; |
---|
| 78 | |
---|
| 79 | /** wether scheme can predict numeric attributes (array index is attribute type of class) */ |
---|
| 80 | protected boolean[] m_NumericPredictors; |
---|
| 81 | |
---|
| 82 | /** wether scheme can predict string attributes (array index is attribute type of class) */ |
---|
| 83 | protected boolean[] m_StringPredictors; |
---|
| 84 | |
---|
| 85 | /** wether scheme can predict date attributes (array index is attribute type of class) */ |
---|
| 86 | protected boolean[] m_DatePredictors; |
---|
| 87 | |
---|
| 88 | /** wether scheme can predict relational attributes (array index is attribute type of class) */ |
---|
| 89 | protected boolean[] m_RelationalPredictors; |
---|
| 90 | |
---|
| 91 | /** whether scheme handles missing values */ |
---|
| 92 | protected boolean[] m_handleMissingPredictors; |
---|
| 93 | |
---|
| 94 | /** whether scheme handles class with only missing values */ |
---|
| 95 | protected boolean[] m_handleMissingClass; |
---|
| 96 | |
---|
| 97 | /** whether scheme handles class as first attribute */ |
---|
| 98 | protected boolean[] m_handleClassAsFirstAttribute; |
---|
| 99 | |
---|
| 100 | /** whether scheme handles class as second attribute */ |
---|
| 101 | protected boolean[] m_handleClassAsSecondAttribute; |
---|
| 102 | |
---|
| 103 | /** the results of the regression tests */ |
---|
| 104 | protected String[] m_RegressionResults; |
---|
| 105 | |
---|
| 106 | /** the OptionHandler tester */ |
---|
| 107 | protected CheckOptionHandler m_OptionTester; |
---|
| 108 | |
---|
| 109 | /** for testing GOE stuff */ |
---|
| 110 | protected CheckGOE m_GOETester; |
---|
| 111 | |
---|
| 112 | /** |
---|
| 113 | * Constructs the <code>AbstractAttributeSelectionTest</code>. Called by subclasses. |
---|
| 114 | * |
---|
| 115 | * @param name the name of the test class |
---|
| 116 | */ |
---|
| 117 | public AbstractAttributeSelectionTest(String name) { |
---|
| 118 | super(name); |
---|
| 119 | } |
---|
| 120 | |
---|
| 121 | /** |
---|
| 122 | * returns a custom PostProcessor for the CheckAttributeSelection datasets, |
---|
| 123 | * currently only null. |
---|
| 124 | * |
---|
| 125 | * @return a custom PostProcessor, if necessary |
---|
| 126 | * @see PostProcessor |
---|
| 127 | */ |
---|
| 128 | protected PostProcessor getPostProcessor() { |
---|
| 129 | return null; |
---|
| 130 | } |
---|
| 131 | |
---|
| 132 | /** |
---|
| 133 | * configures the CheckAttributeSelection instance used throughout the tests |
---|
| 134 | * |
---|
| 135 | * @return the fully configured CheckAttributeSelection instance used for testing |
---|
| 136 | */ |
---|
| 137 | protected CheckAttributeSelection getTester() { |
---|
| 138 | CheckAttributeSelection result; |
---|
| 139 | |
---|
| 140 | result = new CheckAttributeSelection(); |
---|
| 141 | result.setSilent(true); |
---|
| 142 | result.setSearch(m_Search); |
---|
| 143 | result.setEvaluator(m_Evaluator); |
---|
| 144 | result.setNumInstances(20); |
---|
| 145 | result.setDebug(DEBUG); |
---|
| 146 | result.setPostProcessor(getPostProcessor()); |
---|
| 147 | |
---|
| 148 | return result; |
---|
| 149 | } |
---|
| 150 | |
---|
| 151 | /** |
---|
| 152 | * Configures the CheckOptionHandler uses for testing the optionhandling. |
---|
| 153 | * Sets the scheme to test. |
---|
| 154 | * |
---|
| 155 | * @return the fully configured CheckOptionHandler |
---|
| 156 | */ |
---|
| 157 | protected CheckOptionHandler getOptionTester() { |
---|
| 158 | CheckOptionHandler result; |
---|
| 159 | |
---|
| 160 | result = new CheckOptionHandler(); |
---|
| 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 | * |
---|
| 171 | * @return the fully configured CheckGOE |
---|
| 172 | */ |
---|
| 173 | protected CheckGOE getGOETester() { |
---|
| 174 | CheckGOE result; |
---|
| 175 | |
---|
| 176 | result = new CheckGOE(); |
---|
| 177 | result.setObject(null); |
---|
| 178 | result.setSilent(true); |
---|
| 179 | |
---|
| 180 | return result; |
---|
| 181 | } |
---|
| 182 | |
---|
| 183 | /** |
---|
| 184 | * Called by JUnit before each test method. This implementation creates |
---|
| 185 | * the default scheme to test and loads a test set of Instances. |
---|
| 186 | * |
---|
| 187 | * @exception Exception if an error occurs reading the example instances. |
---|
| 188 | */ |
---|
| 189 | protected void setUp() throws Exception { |
---|
| 190 | m_Search = getSearch(); |
---|
| 191 | m_Evaluator = getEvaluator(); |
---|
| 192 | m_Tester = getTester(); |
---|
| 193 | m_OptionTester = getOptionTester(); |
---|
| 194 | m_GOETester = getGOETester(); |
---|
| 195 | |
---|
| 196 | m_weightedInstancesHandler = m_Tester.weightedInstancesHandler()[0]; |
---|
| 197 | m_multiInstanceHandler = m_Tester.multiInstanceHandler()[0]; |
---|
| 198 | m_NominalPredictors = new boolean[LAST_CLASSTYPE + 1]; |
---|
| 199 | m_NumericPredictors = new boolean[LAST_CLASSTYPE + 1]; |
---|
| 200 | m_StringPredictors = new boolean[LAST_CLASSTYPE + 1]; |
---|
| 201 | m_DatePredictors = new boolean[LAST_CLASSTYPE + 1]; |
---|
| 202 | m_RelationalPredictors = new boolean[LAST_CLASSTYPE + 1]; |
---|
| 203 | m_handleMissingPredictors = new boolean[LAST_CLASSTYPE + 1]; |
---|
| 204 | m_handleMissingClass = new boolean[LAST_CLASSTYPE + 1]; |
---|
| 205 | m_handleClassAsFirstAttribute = new boolean[LAST_CLASSTYPE + 1]; |
---|
| 206 | m_handleClassAsSecondAttribute = new boolean[LAST_CLASSTYPE + 1]; |
---|
| 207 | m_RegressionResults = new String[LAST_CLASSTYPE + 1]; |
---|
| 208 | m_NClasses = 4; |
---|
| 209 | |
---|
| 210 | // initialize attributes |
---|
| 211 | checkAttributes(true, false, false, false, false, false); |
---|
| 212 | checkAttributes(false, true, false, false, false, false); |
---|
| 213 | checkAttributes(false, false, true, false, false, false); |
---|
| 214 | checkAttributes(false, false, false, true, false, false); |
---|
| 215 | checkAttributes(false, false, false, false, true, false); |
---|
| 216 | |
---|
| 217 | // initialize missing values handling |
---|
| 218 | for (int i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
---|
| 219 | // does the scheme support this type of class at all? |
---|
| 220 | if (!canPredict(i)) |
---|
| 221 | continue; |
---|
| 222 | |
---|
| 223 | // 20% missing |
---|
| 224 | m_handleMissingPredictors[i] = checkMissingPredictors(i, 20, false); |
---|
| 225 | m_handleMissingClass[i] = checkMissingClass(i, 20, false); |
---|
| 226 | } |
---|
| 227 | } |
---|
| 228 | |
---|
| 229 | /** Called by JUnit after each test method */ |
---|
| 230 | protected void tearDown() { |
---|
| 231 | m_Search = null; |
---|
| 232 | m_Evaluator = 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 search scheme. |
---|
| 253 | * |
---|
| 254 | * @return a suitably configured <code>ASSearch</code> value |
---|
| 255 | */ |
---|
| 256 | public abstract ASSearch getSearch(); |
---|
| 257 | |
---|
| 258 | /** |
---|
| 259 | * Used to create an instance of a specific evaluator. |
---|
| 260 | * |
---|
| 261 | * @return a suitably configured <code>ASEvaluation</code> value |
---|
| 262 | */ |
---|
| 263 | public abstract ASEvaluation getEvaluator(); |
---|
| 264 | |
---|
| 265 | /** |
---|
| 266 | * checks whether at least one attribute type can be handled with the |
---|
| 267 | * given class type |
---|
| 268 | * |
---|
| 269 | * @param type the class type to check for |
---|
| 270 | * @return true if at least one attribute type can be predicted with |
---|
| 271 | * the given class |
---|
| 272 | */ |
---|
| 273 | protected boolean canPredict(int type) { |
---|
| 274 | return m_NominalPredictors[type] |
---|
| 275 | || m_NumericPredictors[type] |
---|
| 276 | || m_StringPredictors[type] |
---|
| 277 | || m_DatePredictors[type] |
---|
| 278 | || m_RelationalPredictors[type]; |
---|
| 279 | } |
---|
| 280 | |
---|
| 281 | /** |
---|
| 282 | * returns a string for the class type |
---|
| 283 | * |
---|
| 284 | * @param type the class type |
---|
| 285 | * @return the class type as string |
---|
| 286 | */ |
---|
| 287 | protected String getClassTypeString(int type) { |
---|
| 288 | return CheckAttributeSelection.attributeTypeToString(type); |
---|
| 289 | } |
---|
| 290 | |
---|
| 291 | /** |
---|
| 292 | * tests whether the scheme can handle certain attributes and if not, |
---|
| 293 | * if the exception is OK |
---|
| 294 | * |
---|
| 295 | * @param nom to check for nominal attributes |
---|
| 296 | * @param num to check for numeric attributes |
---|
| 297 | * @param str to check for string attributes |
---|
| 298 | * @param dat to check for date attributes |
---|
| 299 | * @param rel to check for relational attributes |
---|
| 300 | * @param allowFail whether a junit fail can be executed |
---|
| 301 | * @see CheckAttributeSelection#canPredict(boolean, boolean, boolean, boolean, boolean, boolean, int) |
---|
| 302 | * @see CheckAttributeSelection#testsPerClassType(int, boolean, boolean) |
---|
| 303 | */ |
---|
| 304 | protected void checkAttributes(boolean nom, boolean num, boolean str, |
---|
| 305 | boolean dat, boolean rel, |
---|
| 306 | boolean allowFail) { |
---|
| 307 | boolean[] result; |
---|
| 308 | String att; |
---|
| 309 | int i; |
---|
| 310 | |
---|
| 311 | // determine text for type of attributes |
---|
| 312 | att = ""; |
---|
| 313 | if (nom) |
---|
| 314 | att = "nominal"; |
---|
| 315 | else if (num) |
---|
| 316 | att = "numeric"; |
---|
| 317 | else if (str) |
---|
| 318 | att = "string"; |
---|
| 319 | else if (dat) |
---|
| 320 | att = "date"; |
---|
| 321 | else if (rel) |
---|
| 322 | att = "relational"; |
---|
| 323 | |
---|
| 324 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
---|
| 325 | result = m_Tester.canPredict(nom, num, str, dat, rel, m_multiInstanceHandler, i); |
---|
| 326 | if (nom) |
---|
| 327 | m_NominalPredictors[i] = result[0]; |
---|
| 328 | else if (num) |
---|
| 329 | m_NumericPredictors[i] = result[0]; |
---|
| 330 | else if (str) |
---|
| 331 | m_StringPredictors[i] = result[0]; |
---|
| 332 | else if (dat) |
---|
| 333 | m_DatePredictors[i] = result[0]; |
---|
| 334 | else if (rel) |
---|
| 335 | m_RelationalPredictors[i] = result[0]; |
---|
| 336 | |
---|
| 337 | if (!result[0] && !result[1] && allowFail) |
---|
| 338 | fail("Error handling " + att + " attributes (" + getClassTypeString(i) |
---|
| 339 | + " class)!"); |
---|
| 340 | } |
---|
| 341 | } |
---|
| 342 | |
---|
| 343 | /** |
---|
| 344 | * tests whether the scheme can handle different types of attributes and |
---|
| 345 | * if not, if the exception is OK |
---|
| 346 | * |
---|
| 347 | * @see #checkAttributes(boolean, boolean, boolean, boolean, boolean, boolean) |
---|
| 348 | */ |
---|
| 349 | public void testAttributes() { |
---|
| 350 | // nominal |
---|
| 351 | checkAttributes(true, false, false, false, false, true); |
---|
| 352 | // numeric |
---|
| 353 | checkAttributes(false, true, false, false, false, true); |
---|
| 354 | // string |
---|
| 355 | checkAttributes(false, false, true, false, false, true); |
---|
| 356 | // date |
---|
| 357 | checkAttributes(false, false, false, true, false, true); |
---|
| 358 | // relational |
---|
| 359 | if (!m_multiInstanceHandler) |
---|
| 360 | checkAttributes(false, false, false, false, true, true); |
---|
| 361 | } |
---|
| 362 | |
---|
| 363 | /** |
---|
| 364 | * tests whether the scheme declares a serialVersionUID. |
---|
| 365 | */ |
---|
| 366 | public void testSerialVersionUID() { |
---|
| 367 | boolean[] result; |
---|
| 368 | |
---|
| 369 | result = m_Tester.declaresSerialVersionUID(); |
---|
| 370 | |
---|
| 371 | if (!result[0]) |
---|
| 372 | fail("Doesn't declare serialVersionUID!"); |
---|
| 373 | } |
---|
| 374 | |
---|
| 375 | /** |
---|
| 376 | * tests whether the scheme handles instance weights correctly |
---|
| 377 | * |
---|
| 378 | * @see CheckAttributeSelection#instanceWeights(boolean, boolean, boolean, boolean, boolean, boolean, int) |
---|
| 379 | * @see CheckAttributeSelection#testsPerClassType(int, boolean, boolean) |
---|
| 380 | */ |
---|
| 381 | public void testInstanceWeights() { |
---|
| 382 | boolean[] result; |
---|
| 383 | int i; |
---|
| 384 | |
---|
| 385 | if (m_weightedInstancesHandler) { |
---|
| 386 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
---|
| 387 | // does the scheme support this type of class at all? |
---|
| 388 | if (!canPredict(i)) |
---|
| 389 | continue; |
---|
| 390 | |
---|
| 391 | result = m_Tester.instanceWeights( |
---|
| 392 | m_NominalPredictors[i], |
---|
| 393 | m_NumericPredictors[i], |
---|
| 394 | m_StringPredictors[i], |
---|
| 395 | m_DatePredictors[i], |
---|
| 396 | m_RelationalPredictors[i], |
---|
| 397 | m_multiInstanceHandler, |
---|
| 398 | i); |
---|
| 399 | |
---|
| 400 | if (!result[0]) |
---|
| 401 | System.err.println("Error handling instance weights (" + getClassTypeString(i) |
---|
| 402 | + " class)!"); |
---|
| 403 | } |
---|
| 404 | } |
---|
| 405 | } |
---|
| 406 | |
---|
| 407 | /** |
---|
| 408 | * tests whether scheme handles N classes |
---|
| 409 | * |
---|
| 410 | * @see CheckAttributeSelection#canHandleNClasses(boolean, boolean, boolean, boolean, boolean, boolean, int) |
---|
| 411 | * @see CheckAttributeSelection#testsPerClassType(int, boolean, boolean) |
---|
| 412 | * @see #m_NClasses |
---|
| 413 | */ |
---|
| 414 | public void testNClasses() { |
---|
| 415 | boolean[] result; |
---|
| 416 | |
---|
| 417 | if (!canPredict(Attribute.NOMINAL)) |
---|
| 418 | return; |
---|
| 419 | |
---|
| 420 | result = m_Tester.canHandleNClasses( |
---|
| 421 | m_NominalPredictors[Attribute.NOMINAL], |
---|
| 422 | m_NumericPredictors[Attribute.NOMINAL], |
---|
| 423 | m_StringPredictors[Attribute.NOMINAL], |
---|
| 424 | m_DatePredictors[Attribute.NOMINAL], |
---|
| 425 | m_RelationalPredictors[Attribute.NOMINAL], |
---|
| 426 | m_multiInstanceHandler, |
---|
| 427 | m_NClasses); |
---|
| 428 | |
---|
| 429 | if (!result[0] && !result[1]) |
---|
| 430 | fail("Error handling " + m_NClasses + " classes!"); |
---|
| 431 | } |
---|
| 432 | |
---|
| 433 | /** |
---|
| 434 | * checks whether the scheme can handle the class attribute at a given |
---|
| 435 | * position (0-based index, -1 means last). |
---|
| 436 | * |
---|
| 437 | * @param type the class type |
---|
| 438 | * @param position the position of the class attribute (0-based, -1 means last) |
---|
| 439 | * @return true if the scheme can handle it |
---|
| 440 | */ |
---|
| 441 | protected boolean checkClassAsNthAttribute(int type, int position) { |
---|
| 442 | boolean[] result; |
---|
| 443 | String indexStr; |
---|
| 444 | |
---|
| 445 | result = m_Tester.canHandleClassAsNthAttribute( |
---|
| 446 | m_NominalPredictors[type], |
---|
| 447 | m_NumericPredictors[type], |
---|
| 448 | m_StringPredictors[type], |
---|
| 449 | m_DatePredictors[type], |
---|
| 450 | m_RelationalPredictors[type], |
---|
| 451 | m_multiInstanceHandler, |
---|
| 452 | type, |
---|
| 453 | position); |
---|
| 454 | |
---|
| 455 | if (position == -1) |
---|
| 456 | indexStr = "last"; |
---|
| 457 | else |
---|
| 458 | indexStr = (position + 1) + "."; |
---|
| 459 | |
---|
| 460 | if (!result[0] && !result[1]) |
---|
| 461 | fail("Error handling class as " + indexStr + " attribute (" |
---|
| 462 | + getClassTypeString(type) + " class)!"); |
---|
| 463 | |
---|
| 464 | return result[0]; |
---|
| 465 | } |
---|
| 466 | |
---|
| 467 | /** |
---|
| 468 | * Tests whether the scheme can handle class attributes as Nth |
---|
| 469 | * attribute. In case of multi-instance handler it performs no tests, |
---|
| 470 | * since the multi-instance data has a fixed format (bagID,bag,class). |
---|
| 471 | * |
---|
| 472 | * @see CheckAttributeSelection#canHandleClassAsNthAttribute(boolean, boolean, boolean, boolean, boolean, boolean, int, int) |
---|
| 473 | * @see CheckAttributeSelection#testsPerClassType(int, boolean, boolean) |
---|
| 474 | */ |
---|
| 475 | public void testClassAsNthAttribute() { |
---|
| 476 | int i; |
---|
| 477 | |
---|
| 478 | // multi-Instance data has fixed format! |
---|
| 479 | if (m_multiInstanceHandler) |
---|
| 480 | return; |
---|
| 481 | |
---|
| 482 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
---|
| 483 | // does the scheme support this type of class at all? |
---|
| 484 | if (!canPredict(i)) |
---|
| 485 | continue; |
---|
| 486 | |
---|
| 487 | // first attribute |
---|
| 488 | m_handleClassAsFirstAttribute[i] = checkClassAsNthAttribute(i, 0); |
---|
| 489 | |
---|
| 490 | // second attribute |
---|
| 491 | m_handleClassAsSecondAttribute[i] = checkClassAsNthAttribute(i, 1); |
---|
| 492 | } |
---|
| 493 | } |
---|
| 494 | |
---|
| 495 | /** |
---|
| 496 | * tests whether the scheme can handle zero training instances |
---|
| 497 | * |
---|
| 498 | * @see CheckAttributeSelection#canHandleZeroTraining(boolean, boolean, boolean, boolean, boolean, boolean, int) |
---|
| 499 | * @see CheckAttributeSelection#testsPerClassType(int, boolean, boolean) |
---|
| 500 | */ |
---|
| 501 | public void testZeroTraining() { |
---|
| 502 | boolean[] result; |
---|
| 503 | int i; |
---|
| 504 | |
---|
| 505 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
---|
| 506 | // does the scheme support this type of class at all? |
---|
| 507 | if (!canPredict(i)) |
---|
| 508 | continue; |
---|
| 509 | |
---|
| 510 | result = m_Tester.canHandleZeroTraining( |
---|
| 511 | m_NominalPredictors[i], |
---|
| 512 | m_NumericPredictors[i], |
---|
| 513 | m_StringPredictors[i], |
---|
| 514 | m_DatePredictors[i], |
---|
| 515 | m_RelationalPredictors[i], |
---|
| 516 | m_multiInstanceHandler, |
---|
| 517 | i); |
---|
| 518 | |
---|
| 519 | if (!result[0] && !result[1]) |
---|
| 520 | fail("Error handling zero training instances (" + getClassTypeString(i) |
---|
| 521 | + " class)!"); |
---|
| 522 | } |
---|
| 523 | } |
---|
| 524 | |
---|
| 525 | /** |
---|
| 526 | * checks whether the scheme can handle the given percentage of |
---|
| 527 | * missing predictors |
---|
| 528 | * |
---|
| 529 | * @param type the class type |
---|
| 530 | * @param percent the percentage of missing predictors |
---|
| 531 | * @param allowFail if true a fail statement may be executed |
---|
| 532 | * @return true if the scheme can handle it |
---|
| 533 | */ |
---|
| 534 | protected boolean checkMissingPredictors(int type, int percent, boolean allowFail) { |
---|
| 535 | boolean[] result; |
---|
| 536 | |
---|
| 537 | result = m_Tester.canHandleMissing( |
---|
| 538 | m_NominalPredictors[type], |
---|
| 539 | m_NumericPredictors[type], |
---|
| 540 | m_StringPredictors[type], |
---|
| 541 | m_DatePredictors[type], |
---|
| 542 | m_RelationalPredictors[type], |
---|
| 543 | m_multiInstanceHandler, |
---|
| 544 | type, |
---|
| 545 | true, |
---|
| 546 | false, |
---|
| 547 | percent); |
---|
| 548 | |
---|
| 549 | if (allowFail) { |
---|
| 550 | if (!result[0] && !result[1]) |
---|
| 551 | fail("Error handling " + percent + "% missing predictors (" |
---|
| 552 | + getClassTypeString(type) + " class)!"); |
---|
| 553 | } |
---|
| 554 | |
---|
| 555 | return result[0]; |
---|
| 556 | } |
---|
| 557 | |
---|
| 558 | /** |
---|
| 559 | * tests whether the scheme can handle missing predictors (20% and 100%) |
---|
| 560 | * |
---|
| 561 | * @see CheckAttributeSelection#canHandleMissing(boolean, boolean, boolean, boolean, boolean, boolean, int, boolean, boolean, int) |
---|
| 562 | * @see CheckAttributeSelection#testsPerClassType(int, boolean, boolean) |
---|
| 563 | */ |
---|
| 564 | public void testMissingPredictors() { |
---|
| 565 | int i; |
---|
| 566 | |
---|
| 567 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
---|
| 568 | // does the scheme support this type of class at all? |
---|
| 569 | if (!canPredict(i)) |
---|
| 570 | continue; |
---|
| 571 | |
---|
| 572 | // 20% missing |
---|
| 573 | checkMissingPredictors(i, 20, true); |
---|
| 574 | |
---|
| 575 | // 100% missing |
---|
| 576 | if (m_handleMissingPredictors[i]) |
---|
| 577 | checkMissingPredictors(i, 100, true); |
---|
| 578 | } |
---|
| 579 | } |
---|
| 580 | |
---|
| 581 | /** |
---|
| 582 | * checks whether the scheme can handle the given percentage of |
---|
| 583 | * missing class labels |
---|
| 584 | * |
---|
| 585 | * @param type the class type |
---|
| 586 | * @param percent the percentage of missing class labels |
---|
| 587 | * @param allowFail if true a fail statement may be executed |
---|
| 588 | * @return true if the scheme can handle it |
---|
| 589 | */ |
---|
| 590 | protected boolean checkMissingClass(int type, int percent, boolean allowFail) { |
---|
| 591 | boolean[] result; |
---|
| 592 | |
---|
| 593 | result = m_Tester.canHandleMissing( |
---|
| 594 | m_NominalPredictors[type], |
---|
| 595 | m_NumericPredictors[type], |
---|
| 596 | m_StringPredictors[type], |
---|
| 597 | m_DatePredictors[type], |
---|
| 598 | m_RelationalPredictors[type], |
---|
| 599 | m_multiInstanceHandler, |
---|
| 600 | type, |
---|
| 601 | false, |
---|
| 602 | true, |
---|
| 603 | percent); |
---|
| 604 | |
---|
| 605 | if (allowFail) { |
---|
| 606 | if (!result[0] && !result[1]) |
---|
| 607 | fail("Error handling " + percent + "% missing class labels (" |
---|
| 608 | + getClassTypeString(type) + " class)!"); |
---|
| 609 | } |
---|
| 610 | |
---|
| 611 | return result[0]; |
---|
| 612 | } |
---|
| 613 | |
---|
| 614 | /** |
---|
| 615 | * tests whether the scheme can handle missing class values (20% and |
---|
| 616 | * 100%) |
---|
| 617 | * |
---|
| 618 | * @see CheckAttributeSelection#canHandleMissing(boolean, boolean, boolean, boolean, boolean, boolean, int, boolean, boolean, int) |
---|
| 619 | * @see CheckAttributeSelection#testsPerClassType(int, boolean, boolean) |
---|
| 620 | */ |
---|
| 621 | public void testMissingClass() { |
---|
| 622 | int i; |
---|
| 623 | |
---|
| 624 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
---|
| 625 | // does the scheme support this type of class at all? |
---|
| 626 | if (!canPredict(i)) |
---|
| 627 | continue; |
---|
| 628 | |
---|
| 629 | // 20% missing |
---|
| 630 | checkMissingClass(i, 20, true); |
---|
| 631 | |
---|
| 632 | // 100% missing |
---|
| 633 | if (m_handleMissingClass[i]) |
---|
| 634 | checkMissingClass(i, 100, true); |
---|
| 635 | } |
---|
| 636 | } |
---|
| 637 | |
---|
| 638 | /** |
---|
| 639 | * tests whether the scheme correctly initializes in the |
---|
| 640 | * search method |
---|
| 641 | * |
---|
| 642 | * @see CheckAttributeSelection#correctSearchInitialisation(boolean, boolean, boolean, boolean, boolean, boolean, int) |
---|
| 643 | * @see CheckAttributeSelection#testsPerClassType(int, boolean, boolean) |
---|
| 644 | */ |
---|
| 645 | public void testBuildInitialization() { |
---|
| 646 | boolean[] result; |
---|
| 647 | int i; |
---|
| 648 | |
---|
| 649 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
---|
| 650 | // does the scheme support this type of class at all? |
---|
| 651 | if (!canPredict(i)) |
---|
| 652 | continue; |
---|
| 653 | |
---|
| 654 | result = m_Tester.correctSearchInitialisation( |
---|
| 655 | m_NominalPredictors[i], |
---|
| 656 | m_NumericPredictors[i], |
---|
| 657 | m_StringPredictors[i], |
---|
| 658 | m_DatePredictors[i], |
---|
| 659 | m_RelationalPredictors[i], |
---|
| 660 | m_multiInstanceHandler, |
---|
| 661 | i); |
---|
| 662 | |
---|
| 663 | if (!result[0] && !result[1]) |
---|
| 664 | fail("Incorrect build initialization (" + getClassTypeString(i) |
---|
| 665 | + " class)!"); |
---|
| 666 | } |
---|
| 667 | } |
---|
| 668 | |
---|
| 669 | /** |
---|
| 670 | * tests whether the scheme alters the training set during training. |
---|
| 671 | * |
---|
| 672 | * @see CheckAttributeSelection#datasetIntegrity(boolean, boolean, boolean, boolean, boolean, boolean, int, boolean, boolean) |
---|
| 673 | * @see CheckAttributeSelection#testsPerClassType(int, boolean, boolean) |
---|
| 674 | */ |
---|
| 675 | public void testDatasetIntegrity() { |
---|
| 676 | boolean[] result; |
---|
| 677 | int i; |
---|
| 678 | |
---|
| 679 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
---|
| 680 | // does the scheme support this type of class at all? |
---|
| 681 | if (!canPredict(i)) |
---|
| 682 | continue; |
---|
| 683 | |
---|
| 684 | result = m_Tester.datasetIntegrity( |
---|
| 685 | m_NominalPredictors[i], |
---|
| 686 | m_NumericPredictors[i], |
---|
| 687 | m_StringPredictors[i], |
---|
| 688 | m_DatePredictors[i], |
---|
| 689 | m_RelationalPredictors[i], |
---|
| 690 | m_multiInstanceHandler, |
---|
| 691 | i, |
---|
| 692 | m_handleMissingPredictors[i], |
---|
| 693 | m_handleMissingClass[i]); |
---|
| 694 | |
---|
| 695 | if (!result[0] && !result[1]) |
---|
| 696 | fail("Dataset is altered during training (" |
---|
| 697 | + getClassTypeString(i) + " class)!"); |
---|
| 698 | } |
---|
| 699 | } |
---|
| 700 | |
---|
| 701 | /** |
---|
| 702 | * Builds a model using the current scheme using the given data. |
---|
| 703 | * |
---|
| 704 | * @param data the instances to test the selection scheme on |
---|
| 705 | * @return a string containing the results. |
---|
| 706 | */ |
---|
| 707 | protected String useScheme(Instances data) throws Exception { |
---|
| 708 | AttributeSelection attsel = null; |
---|
| 709 | |
---|
| 710 | try { |
---|
| 711 | attsel = new AttributeSelection(); |
---|
| 712 | attsel.setSearch(m_Search); |
---|
| 713 | attsel.setEvaluator(m_Evaluator); |
---|
| 714 | attsel.setSeed(42); |
---|
| 715 | } |
---|
| 716 | catch (Exception e) { |
---|
| 717 | e.printStackTrace(); |
---|
| 718 | fail("Problem setting up attribute selection: " + e); |
---|
| 719 | } |
---|
| 720 | |
---|
| 721 | attsel.SelectAttributes(data); |
---|
| 722 | |
---|
| 723 | return attsel.toResultsString(); |
---|
| 724 | } |
---|
| 725 | |
---|
| 726 | /** |
---|
| 727 | * Provides a hook for derived classes to further modify the data. Currently, |
---|
| 728 | * the data is just passed through. |
---|
| 729 | * |
---|
| 730 | * @param data the data to process |
---|
| 731 | * @return the processed data |
---|
| 732 | */ |
---|
| 733 | protected Instances process(Instances data) { |
---|
| 734 | return data; |
---|
| 735 | } |
---|
| 736 | |
---|
| 737 | /** |
---|
| 738 | * Runs a regression test -- this checks that the output of the tested |
---|
| 739 | * object matches that in a reference version. When this test is |
---|
| 740 | * run without any pre-existing reference output, the reference version |
---|
| 741 | * is created. |
---|
| 742 | */ |
---|
| 743 | public void testRegression() throws Exception { |
---|
| 744 | int i; |
---|
| 745 | boolean succeeded; |
---|
| 746 | Regression reg; |
---|
| 747 | Instances train; |
---|
| 748 | |
---|
| 749 | // don't bother if not working correctly |
---|
| 750 | if (m_Tester.hasClasspathProblems()) |
---|
| 751 | return; |
---|
| 752 | |
---|
| 753 | reg = new Regression(this.getClass()); |
---|
| 754 | succeeded = false; |
---|
| 755 | train = null; |
---|
| 756 | |
---|
| 757 | for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) { |
---|
| 758 | // does the scheme support this type of class at all? |
---|
| 759 | if (!canPredict(i)) |
---|
| 760 | continue; |
---|
| 761 | |
---|
| 762 | train = m_Tester.makeTestDataset( |
---|
| 763 | 42, m_Tester.getNumInstances(), |
---|
| 764 | m_NominalPredictors[i] ? 2 : 0, |
---|
| 765 | m_NumericPredictors[i] ? 1 : 0, |
---|
| 766 | m_StringPredictors[i] ? 1 : 0, |
---|
| 767 | m_DatePredictors[i] ? 1 : 0, |
---|
| 768 | m_RelationalPredictors[i] ? 1 : 0, |
---|
| 769 | 2, |
---|
| 770 | i, |
---|
| 771 | m_multiInstanceHandler); |
---|
| 772 | |
---|
| 773 | try { |
---|
| 774 | m_RegressionResults[i] = useScheme(train); |
---|
| 775 | succeeded = true; |
---|
| 776 | reg.println(m_RegressionResults[i]); |
---|
| 777 | } |
---|
| 778 | catch (Exception e) { |
---|
| 779 | String msg = e.getMessage().toLowerCase(); |
---|
| 780 | if (msg.indexOf("not in classpath") > -1) |
---|
| 781 | return; |
---|
| 782 | |
---|
| 783 | m_RegressionResults[i] = null; |
---|
| 784 | } |
---|
| 785 | } |
---|
| 786 | |
---|
| 787 | if (!succeeded) { |
---|
| 788 | fail("Problem during regression testing: no successful predictions for any class type"); |
---|
| 789 | } |
---|
| 790 | |
---|
| 791 | try { |
---|
| 792 | String diff = reg.diff(); |
---|
| 793 | if (diff == null) { |
---|
| 794 | System.err.println("Warning: No reference available, creating."); |
---|
| 795 | } else if (!diff.equals("")) { |
---|
| 796 | fail("Regression test failed. Difference:\n" + diff); |
---|
| 797 | } |
---|
| 798 | } |
---|
| 799 | catch (java.io.IOException ex) { |
---|
| 800 | fail("Problem during regression testing.\n" + ex); |
---|
| 801 | } |
---|
| 802 | } |
---|
| 803 | |
---|
| 804 | /** |
---|
| 805 | * tests the listing of the options |
---|
| 806 | */ |
---|
| 807 | public void testListOptions() { |
---|
| 808 | if (m_OptionTester.getOptionHandler() != null) { |
---|
| 809 | if (!m_OptionTester.checkListOptions()) |
---|
| 810 | fail("Options cannot be listed via listOptions."); |
---|
| 811 | } |
---|
| 812 | } |
---|
| 813 | |
---|
| 814 | /** |
---|
| 815 | * tests the setting of the options |
---|
| 816 | */ |
---|
| 817 | public void testSetOptions() { |
---|
| 818 | if (m_OptionTester.getOptionHandler() != null) { |
---|
| 819 | if (!m_OptionTester.checkSetOptions()) |
---|
| 820 | fail("setOptions method failed."); |
---|
| 821 | } |
---|
| 822 | } |
---|
| 823 | |
---|
| 824 | /** |
---|
| 825 | * tests whether the default settings are processed correctly |
---|
| 826 | */ |
---|
| 827 | public void testDefaultOptions() { |
---|
| 828 | if (m_OptionTester.getOptionHandler() != null) { |
---|
| 829 | if (!m_OptionTester.checkDefaultOptions()) |
---|
| 830 | fail("Default options were not processed correctly."); |
---|
| 831 | } |
---|
| 832 | } |
---|
| 833 | |
---|
| 834 | /** |
---|
| 835 | * tests whether there are any remaining options |
---|
| 836 | */ |
---|
| 837 | public void testRemainingOptions() { |
---|
| 838 | if (m_OptionTester.getOptionHandler() != null) { |
---|
| 839 | if (!m_OptionTester.checkRemainingOptions()) |
---|
| 840 | fail("There were 'left-over' options."); |
---|
| 841 | } |
---|
| 842 | } |
---|
| 843 | |
---|
| 844 | /** |
---|
| 845 | * tests the whether the user-supplied options stay the same after setting. |
---|
| 846 | * getting, and re-setting again. |
---|
| 847 | * |
---|
| 848 | * @see #getOptionTester() |
---|
| 849 | */ |
---|
| 850 | public void testCanonicalUserOptions() { |
---|
| 851 | if (m_OptionTester.getOptionHandler() != null) { |
---|
| 852 | if (!m_OptionTester.checkCanonicalUserOptions()) |
---|
| 853 | fail("setOptions method failed"); |
---|
| 854 | } |
---|
| 855 | } |
---|
| 856 | |
---|
| 857 | /** |
---|
| 858 | * tests the resetting of the options to the default ones |
---|
| 859 | */ |
---|
| 860 | public void testResettingOptions() { |
---|
| 861 | if (m_OptionTester.getOptionHandler() != null) { |
---|
| 862 | if (!m_OptionTester.checkSetOptions()) |
---|
| 863 | fail("Resetting of options failed"); |
---|
| 864 | } |
---|
| 865 | } |
---|
| 866 | |
---|
| 867 | /** |
---|
| 868 | * tests for a globalInfo method |
---|
| 869 | */ |
---|
| 870 | public void testGlobalInfo() { |
---|
| 871 | if (!m_GOETester.checkGlobalInfo()) |
---|
| 872 | fail("No globalInfo method"); |
---|
| 873 | } |
---|
| 874 | |
---|
| 875 | /** |
---|
| 876 | * tests the tool tips |
---|
| 877 | */ |
---|
| 878 | public void testToolTips() { |
---|
| 879 | if (!m_GOETester.checkToolTips()) |
---|
| 880 | fail("Tool tips inconsistent"); |
---|
| 881 | } |
---|
| 882 | } |
---|