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 | * GeneralRegression.java |
---|
19 | * Copyright (C) 2008 University of Waikato, Hamilton, New Zealand |
---|
20 | * |
---|
21 | */ |
---|
22 | |
---|
23 | package weka.classifiers.pmml.consumer; |
---|
24 | |
---|
25 | import java.io.Serializable; |
---|
26 | import java.util.ArrayList; |
---|
27 | import org.w3c.dom.Element; |
---|
28 | import org.w3c.dom.Node; |
---|
29 | import org.w3c.dom.NodeList; |
---|
30 | |
---|
31 | import weka.core.Attribute; |
---|
32 | import weka.core.Instance; |
---|
33 | import weka.core.Instances; |
---|
34 | import weka.core.RevisionUtils; |
---|
35 | import weka.core.Utils; |
---|
36 | import weka.core.pmml.*; |
---|
37 | |
---|
38 | /** |
---|
39 | * Class implementing import of PMML General Regression model. Can be |
---|
40 | * used as a Weka classifier for prediction (buildClassifier() |
---|
41 | * raises an Exception). |
---|
42 | * |
---|
43 | * @author Mark Hall (mhall{[at]}pentaho{[dot]}com) |
---|
44 | * @version $Revision: 5987 $ |
---|
45 | */ |
---|
46 | public class GeneralRegression extends PMMLClassifier |
---|
47 | implements Serializable { |
---|
48 | |
---|
49 | /** |
---|
50 | * For serialization |
---|
51 | */ |
---|
52 | private static final long serialVersionUID = 2583880411828388959L; |
---|
53 | |
---|
54 | /** |
---|
55 | * Enumerated type for the model type. |
---|
56 | */ |
---|
57 | enum ModelType { |
---|
58 | |
---|
59 | // same type of model |
---|
60 | REGRESSION ("regression"), |
---|
61 | GENERALLINEAR ("generalLinear"), |
---|
62 | MULTINOMIALLOGISTIC ("multinomialLogistic"), |
---|
63 | ORDINALMULTINOMIAL ("ordinalMultinomial"), |
---|
64 | GENERALIZEDLINEAR ("generalizedLinear"); |
---|
65 | |
---|
66 | private final String m_stringVal; |
---|
67 | ModelType(String name) { |
---|
68 | m_stringVal = name; |
---|
69 | } |
---|
70 | |
---|
71 | public String toString() { |
---|
72 | return m_stringVal; |
---|
73 | } |
---|
74 | } |
---|
75 | |
---|
76 | // the model type |
---|
77 | protected ModelType m_modelType = ModelType.REGRESSION; |
---|
78 | |
---|
79 | // the model name (if defined) |
---|
80 | protected String m_modelName; |
---|
81 | |
---|
82 | // the algorithm name (if defined) |
---|
83 | protected String m_algorithmName; |
---|
84 | |
---|
85 | // the function type (regression or classification) |
---|
86 | protected int m_functionType = Regression.RegressionTable.REGRESSION; |
---|
87 | |
---|
88 | /** |
---|
89 | * Enumerated type for the cumulative link function |
---|
90 | * (ordinal multinomial model type only). |
---|
91 | */ |
---|
92 | enum CumulativeLinkFunction { |
---|
93 | NONE ("none") { |
---|
94 | double eval(double value, double offset) { |
---|
95 | return Double.NaN; // no evaluation defined in this case! |
---|
96 | } |
---|
97 | }, |
---|
98 | LOGIT ("logit") { |
---|
99 | double eval(double value, double offset) { |
---|
100 | return 1.0 / (1.0 + Math.exp(-(value + offset))); |
---|
101 | } |
---|
102 | }, |
---|
103 | PROBIT ("probit") { |
---|
104 | double eval(double value, double offset) { |
---|
105 | return weka.core.matrix.Maths.pnorm(value + offset); |
---|
106 | } |
---|
107 | }, |
---|
108 | CLOGLOG ("cloglog") { |
---|
109 | double eval(double value, double offset) { |
---|
110 | return 1.0 - Math.exp(-Math.exp(value + offset)); |
---|
111 | } |
---|
112 | }, |
---|
113 | LOGLOG ("loglog") { |
---|
114 | double eval(double value, double offset) { |
---|
115 | return Math.exp(-Math.exp(-(value + offset))); |
---|
116 | } |
---|
117 | }, |
---|
118 | CAUCHIT ("cauchit") { |
---|
119 | double eval(double value, double offset) { |
---|
120 | return 0.5 + (1.0 / Math.PI) * Math.atan(value + offset); |
---|
121 | } |
---|
122 | }; |
---|
123 | |
---|
124 | /** |
---|
125 | * Evaluation function. |
---|
126 | * |
---|
127 | * @param value the raw response value |
---|
128 | * @param offset the offset to add to the raw value |
---|
129 | * @return the result of the link function |
---|
130 | */ |
---|
131 | abstract double eval(double value, double offset); |
---|
132 | |
---|
133 | private final String m_stringVal; |
---|
134 | |
---|
135 | /** |
---|
136 | * Constructor |
---|
137 | * |
---|
138 | * @param name textual name for this enum |
---|
139 | */ |
---|
140 | CumulativeLinkFunction(String name) { |
---|
141 | m_stringVal = name; |
---|
142 | } |
---|
143 | |
---|
144 | /* (non-Javadoc) |
---|
145 | * @see java.lang.Enum#toString() |
---|
146 | */ |
---|
147 | public String toString() { |
---|
148 | return m_stringVal; |
---|
149 | } |
---|
150 | } |
---|
151 | |
---|
152 | // cumulative link function (ordinal multinomial only) |
---|
153 | protected CumulativeLinkFunction m_cumulativeLinkFunction |
---|
154 | = CumulativeLinkFunction.NONE; |
---|
155 | |
---|
156 | |
---|
157 | /** |
---|
158 | * Enumerated type for the link function (general linear and |
---|
159 | * generalized linear model types only). |
---|
160 | */ |
---|
161 | enum LinkFunction { |
---|
162 | NONE ("none") { |
---|
163 | double eval(double value, double offset, double trials, |
---|
164 | double distParam, double linkParam) { |
---|
165 | return Double.NaN; // no evaluation defined in this case! |
---|
166 | } |
---|
167 | }, |
---|
168 | CLOGLOG ("cloglog") { |
---|
169 | double eval(double value, double offset, double trials, |
---|
170 | double distParam, double linkParam) { |
---|
171 | return (1.0 - Math.exp(-Math.exp(value + offset))) * trials; |
---|
172 | } |
---|
173 | }, |
---|
174 | IDENTITY ("identity") { |
---|
175 | double eval(double value, double offset, double trials, |
---|
176 | double distParam, double linkParam) { |
---|
177 | return (value + offset) * trials; |
---|
178 | } |
---|
179 | }, |
---|
180 | LOG ("log") { |
---|
181 | double eval(double value, double offset, double trials, |
---|
182 | double distParam, double linkParam) { |
---|
183 | return Math.exp(value + offset) * trials; |
---|
184 | } |
---|
185 | }, |
---|
186 | LOGC ("logc") { |
---|
187 | double eval(double value, double offset, double trials, |
---|
188 | double distParam, double linkParam) { |
---|
189 | return (1.0 - Math.exp(value + offset)) * trials; |
---|
190 | } |
---|
191 | }, |
---|
192 | LOGIT ("logit") { |
---|
193 | double eval(double value, double offset, double trials, |
---|
194 | double distParam, double linkParam) { |
---|
195 | return (1.0 / (1.0 + Math.exp(-(value + offset)))) * trials; |
---|
196 | } |
---|
197 | }, |
---|
198 | LOGLOG ("loglog") { |
---|
199 | double eval(double value, double offset, double trials, |
---|
200 | double distParam, double linkParam) { |
---|
201 | return Math.exp(-Math.exp(-(value + offset))) * trials; |
---|
202 | } |
---|
203 | }, |
---|
204 | NEGBIN ("negbin") { |
---|
205 | double eval(double value, double offset, double trials, |
---|
206 | double distParam, double linkParam) { |
---|
207 | return (1.0 / (distParam * (Math.exp(-(value + offset)) - 1.0))) * trials; |
---|
208 | } |
---|
209 | }, |
---|
210 | ODDSPOWER ("oddspower") { |
---|
211 | double eval(double value, double offset, double trials, |
---|
212 | double distParam, double linkParam) { |
---|
213 | return (linkParam < 0.0 || linkParam > 0.0) |
---|
214 | ? (1.0 / (1.0 + Math.pow(1.0 + linkParam * (value + offset), (-1.0 / linkParam)))) * trials |
---|
215 | : (1.0 / (1.0 + Math.exp(-(value + offset)))) * trials; |
---|
216 | } |
---|
217 | }, |
---|
218 | POWER ("power") { |
---|
219 | double eval(double value, double offset, double trials, |
---|
220 | double distParam, double linkParam) { |
---|
221 | return (linkParam < 0.0 || linkParam > 0.0) |
---|
222 | ? Math.pow(value + offset, (1.0 / linkParam)) * trials |
---|
223 | : Math.exp(value + offset) * trials; |
---|
224 | } |
---|
225 | }, |
---|
226 | PROBIT ("probit") { |
---|
227 | double eval(double value, double offset, double trials, |
---|
228 | double distParam, double linkParam) { |
---|
229 | return weka.core.matrix.Maths.pnorm(value + offset) * trials; |
---|
230 | } |
---|
231 | }; |
---|
232 | |
---|
233 | /** |
---|
234 | * Evaluation function. |
---|
235 | * |
---|
236 | * @param value the raw response value |
---|
237 | * @param offset the offset to add to the raw value |
---|
238 | * @param trials the trials value to multiply the result by |
---|
239 | * @param distParam the distribution parameter (negbin only) |
---|
240 | * @param linkParam the link parameter (power and oddspower only) |
---|
241 | * @return the result of the link function |
---|
242 | */ |
---|
243 | abstract double eval(double value, double offset, double trials, |
---|
244 | double distParam, double linkParam); |
---|
245 | |
---|
246 | private final String m_stringVal; |
---|
247 | |
---|
248 | /** |
---|
249 | * Constructor. |
---|
250 | * |
---|
251 | * @param name the textual name of this link function |
---|
252 | */ |
---|
253 | LinkFunction(String name) { |
---|
254 | m_stringVal = name; |
---|
255 | } |
---|
256 | |
---|
257 | /* (non-Javadoc) |
---|
258 | * @see java.lang.Enum#toString() |
---|
259 | */ |
---|
260 | public String toString() { |
---|
261 | return m_stringVal; |
---|
262 | } |
---|
263 | } |
---|
264 | |
---|
265 | // link function (generalLinear model type only) |
---|
266 | protected LinkFunction m_linkFunction = LinkFunction.NONE; |
---|
267 | protected double m_linkParameter = Double.NaN; |
---|
268 | protected String m_trialsVariable; |
---|
269 | protected double m_trialsValue = Double.NaN; |
---|
270 | |
---|
271 | /** |
---|
272 | * Enumerated type for the distribution (general linear |
---|
273 | * and generalized linear model types only). |
---|
274 | */ |
---|
275 | enum Distribution { |
---|
276 | NONE ("none"), |
---|
277 | NORMAL ("normal"), |
---|
278 | BINOMIAL ("binomial"), |
---|
279 | GAMMA ("gamma"), |
---|
280 | INVGAUSSIAN ("igauss"), |
---|
281 | NEGBINOMIAL ("negbin"), |
---|
282 | POISSON ("poisson"); |
---|
283 | |
---|
284 | private final String m_stringVal; |
---|
285 | Distribution(String name) { |
---|
286 | m_stringVal = name; |
---|
287 | } |
---|
288 | |
---|
289 | /* (non-Javadoc) |
---|
290 | * @see java.lang.Enum#toString() |
---|
291 | */ |
---|
292 | public String toString() { |
---|
293 | return m_stringVal; |
---|
294 | } |
---|
295 | } |
---|
296 | |
---|
297 | // generalLinear and generalizedLinear model type only |
---|
298 | protected Distribution m_distribution = Distribution.NORMAL; |
---|
299 | |
---|
300 | // ancillary parameter value for the negative binomial distribution |
---|
301 | protected double m_distParameter = Double.NaN; |
---|
302 | |
---|
303 | // if present, this variable is used during scoring generalizedLinear/generalLinear or |
---|
304 | // ordinalMultinomial models |
---|
305 | protected String m_offsetVariable; |
---|
306 | |
---|
307 | // if present, this variable is used during scoring generalizedLinear/generalLinear or |
---|
308 | // ordinalMultinomial models. It works like a user-specified intercept. |
---|
309 | // At most, only one of offsetVariable or offsetValue may be specified. |
---|
310 | protected double m_offsetValue = Double.NaN; |
---|
311 | |
---|
312 | /** |
---|
313 | * Small inner class to hold the name of a parameter plus |
---|
314 | * its optional descriptive label |
---|
315 | */ |
---|
316 | static class Parameter implements Serializable { |
---|
317 | // ESCA-JAVA0096: |
---|
318 | /** For serialization */ |
---|
319 | // CHECK ME WITH serialver |
---|
320 | private static final long serialVersionUID = 6502780192411755341L; |
---|
321 | |
---|
322 | protected String m_name = null; |
---|
323 | protected String m_label = null; |
---|
324 | } |
---|
325 | |
---|
326 | // List of model parameters |
---|
327 | protected ArrayList<Parameter> m_parameterList = new ArrayList<Parameter>(); |
---|
328 | |
---|
329 | /** |
---|
330 | * Small inner class to hold the name of a factor or covariate, |
---|
331 | * plus the index of the attribute it corresponds to in the |
---|
332 | * mining schema. |
---|
333 | */ |
---|
334 | static class Predictor implements Serializable { |
---|
335 | /** For serialization */ |
---|
336 | // CHECK ME WITH serialver |
---|
337 | private static final long serialVersionUID = 6502780192411755341L; |
---|
338 | |
---|
339 | protected String m_name = null; |
---|
340 | protected int m_miningSchemaIndex = -1; |
---|
341 | |
---|
342 | public String toString() { |
---|
343 | return m_name; |
---|
344 | } |
---|
345 | } |
---|
346 | |
---|
347 | // FactorList |
---|
348 | protected ArrayList<Predictor> m_factorList = new ArrayList<Predictor>(); |
---|
349 | |
---|
350 | // CovariateList |
---|
351 | protected ArrayList<Predictor> m_covariateList = new ArrayList<Predictor>(); |
---|
352 | |
---|
353 | /** |
---|
354 | * Small inner class to hold details on a predictor-to-parameter |
---|
355 | * correlation. |
---|
356 | */ |
---|
357 | static class PPCell implements Serializable { |
---|
358 | /** For serialization */ |
---|
359 | // CHECK ME WITH serialver |
---|
360 | private static final long serialVersionUID = 6502780192411755341L; |
---|
361 | |
---|
362 | protected String m_predictorName = null; |
---|
363 | protected String m_parameterName = null; |
---|
364 | |
---|
365 | // either the exponent of a numeric attribute or the index of |
---|
366 | // a discrete value |
---|
367 | protected double m_value = 0; |
---|
368 | |
---|
369 | // optional. The default is for all target categories to |
---|
370 | // share the same PPMatrix. |
---|
371 | // TO-DO: implement multiple PPMatrixes |
---|
372 | protected String m_targetCategory = null; |
---|
373 | |
---|
374 | } |
---|
375 | |
---|
376 | // PPMatrix (predictor-to-parameter matrix) |
---|
377 | // rows = parameters, columns = predictors (attributes) |
---|
378 | protected PPCell[][] m_ppMatrix; |
---|
379 | |
---|
380 | /** |
---|
381 | * Small inner class to hold a single entry in the |
---|
382 | * ParamMatrix (parameter matrix). |
---|
383 | */ |
---|
384 | static class PCell implements Serializable { |
---|
385 | |
---|
386 | /** For serialization */ |
---|
387 | // CHECK ME WITH serialver |
---|
388 | private static final long serialVersionUID = 6502780192411755341L; |
---|
389 | |
---|
390 | // may be null for numeric target. May also be null if this coefficent |
---|
391 | // applies to all target categories. |
---|
392 | protected String m_targetCategory = null; |
---|
393 | protected String m_parameterName = null; |
---|
394 | // coefficient |
---|
395 | protected double m_beta = 0.0; |
---|
396 | // optional degrees of freedom |
---|
397 | protected int m_df = -1; |
---|
398 | } |
---|
399 | |
---|
400 | // ParamMatrix. rows = target categories (only one if target is numeric), |
---|
401 | // columns = parameters (in order that they occur in the parameter list). |
---|
402 | protected PCell[][] m_paramMatrix; |
---|
403 | |
---|
404 | /** |
---|
405 | * Constructs a GeneralRegression classifier. |
---|
406 | * |
---|
407 | * @param model the Element that holds the model definition |
---|
408 | * @param dataDictionary the data dictionary as a set of Instances |
---|
409 | * @param miningSchema the mining schema |
---|
410 | * @throws Exception if there is a problem constructing the general regression |
---|
411 | * object from the PMML. |
---|
412 | */ |
---|
413 | public GeneralRegression(Element model, Instances dataDictionary, |
---|
414 | MiningSchema miningSchema) throws Exception { |
---|
415 | |
---|
416 | super(dataDictionary, miningSchema); |
---|
417 | |
---|
418 | // get the model type |
---|
419 | String mType = model.getAttribute("modelType"); |
---|
420 | boolean found = false; |
---|
421 | for (ModelType m : ModelType.values()) { |
---|
422 | if (m.toString().equals(mType)) { |
---|
423 | m_modelType = m; |
---|
424 | found = true; |
---|
425 | break; |
---|
426 | } |
---|
427 | } |
---|
428 | if (!found) { |
---|
429 | throw new Exception("[GeneralRegression] unknown model type: " + mType); |
---|
430 | } |
---|
431 | |
---|
432 | if (m_modelType == ModelType.ORDINALMULTINOMIAL) { |
---|
433 | // get the cumulative link function |
---|
434 | String cLink = model.getAttribute("cumulativeLink"); |
---|
435 | found = false; |
---|
436 | for (CumulativeLinkFunction c : CumulativeLinkFunction.values()) { |
---|
437 | if (c.toString().equals(cLink)) { |
---|
438 | m_cumulativeLinkFunction = c; |
---|
439 | found = true; |
---|
440 | break; |
---|
441 | } |
---|
442 | } |
---|
443 | if (!found) { |
---|
444 | throw new Exception("[GeneralRegression] cumulative link function " + cLink); |
---|
445 | } |
---|
446 | } else if (m_modelType == ModelType.GENERALIZEDLINEAR || |
---|
447 | m_modelType == ModelType.GENERALLINEAR) { |
---|
448 | // get the link function |
---|
449 | String link = model.getAttribute("linkFunction"); |
---|
450 | found = false; |
---|
451 | for (LinkFunction l : LinkFunction.values()) { |
---|
452 | if (l.toString().equals(link)) { |
---|
453 | m_linkFunction = l; |
---|
454 | found = true; |
---|
455 | break; |
---|
456 | } |
---|
457 | } |
---|
458 | if (!found) { |
---|
459 | throw new Exception("[GeneralRegression] unknown link function " + link); |
---|
460 | } |
---|
461 | |
---|
462 | // get the link parameter |
---|
463 | String linkP = model.getAttribute("linkParameter"); |
---|
464 | if (linkP != null && linkP.length() > 0) { |
---|
465 | try { |
---|
466 | m_linkParameter = Double.parseDouble(linkP); |
---|
467 | } catch (IllegalArgumentException ex) { |
---|
468 | throw new Exception("[GeneralRegression] unable to parse the link parameter"); |
---|
469 | } |
---|
470 | } |
---|
471 | |
---|
472 | // get the trials variable |
---|
473 | String trials = model.getAttribute("trialsVariable"); |
---|
474 | if (trials != null && trials.length() > 0) { |
---|
475 | m_trialsVariable = trials; |
---|
476 | } |
---|
477 | |
---|
478 | // get the trials value |
---|
479 | String trialsV = model.getAttribute("trialsValue"); |
---|
480 | if (trialsV != null && trialsV.length() > 0) { |
---|
481 | try { |
---|
482 | m_trialsValue = Double.parseDouble(trialsV); |
---|
483 | } catch (IllegalArgumentException ex) { |
---|
484 | throw new Exception("[GeneralRegression] unable to parse the trials value"); |
---|
485 | } |
---|
486 | } |
---|
487 | } |
---|
488 | |
---|
489 | String mName = model.getAttribute("modelName"); |
---|
490 | if (mName != null && mName.length() > 0) { |
---|
491 | m_modelName = mName; |
---|
492 | } |
---|
493 | |
---|
494 | String fName = model.getAttribute("functionName"); |
---|
495 | if (fName.equals("classification")) { |
---|
496 | m_functionType = Regression.RegressionTable.CLASSIFICATION; |
---|
497 | } |
---|
498 | |
---|
499 | String algName = model.getAttribute("algorithmName"); |
---|
500 | if (algName != null && algName.length() > 0) { |
---|
501 | m_algorithmName = algName; |
---|
502 | } |
---|
503 | |
---|
504 | String distribution = model.getAttribute("distribution"); |
---|
505 | if (distribution != null && distribution.length() > 0) { |
---|
506 | found = false; |
---|
507 | for (Distribution d : Distribution.values()) { |
---|
508 | if (d.toString().equals(distribution)) { |
---|
509 | m_distribution = d; |
---|
510 | found = true; |
---|
511 | break; |
---|
512 | } |
---|
513 | } |
---|
514 | if (!found) { |
---|
515 | throw new Exception("[GeneralRegression] unknown distribution type " + distribution); |
---|
516 | } |
---|
517 | } |
---|
518 | |
---|
519 | String distP = model.getAttribute("distParameter"); |
---|
520 | if (distP != null && distP.length() > 0) { |
---|
521 | try { |
---|
522 | m_distParameter = Double.parseDouble(distP); |
---|
523 | } catch (IllegalArgumentException ex) { |
---|
524 | throw new Exception("[GeneralRegression] unable to parse the distribution parameter"); |
---|
525 | } |
---|
526 | } |
---|
527 | |
---|
528 | String offsetV = model.getAttribute("offsetVariable"); |
---|
529 | if (offsetV != null && offsetV.length() > 0) { |
---|
530 | m_offsetVariable = offsetV; |
---|
531 | } |
---|
532 | |
---|
533 | String offsetVal = model.getAttribute("offsetValue"); |
---|
534 | if (offsetVal != null && offsetVal.length() > 0) { |
---|
535 | try { |
---|
536 | m_offsetValue = Double.parseDouble(offsetVal); |
---|
537 | } catch (IllegalArgumentException ex) { |
---|
538 | throw new Exception("[GeneralRegression] unable to parse the offset value"); |
---|
539 | } |
---|
540 | } |
---|
541 | |
---|
542 | // get the parameter list |
---|
543 | readParameterList(model); |
---|
544 | |
---|
545 | // get the factors and covariates |
---|
546 | readFactorsAndCovariates(model, "FactorList"); |
---|
547 | readFactorsAndCovariates(model, "CovariateList"); |
---|
548 | |
---|
549 | // read the PPMatrix |
---|
550 | readPPMatrix(model); |
---|
551 | |
---|
552 | // read the parameter estimates |
---|
553 | readParamMatrix(model); |
---|
554 | } |
---|
555 | |
---|
556 | /** |
---|
557 | * Read the list of parameters. |
---|
558 | * |
---|
559 | * @param model the Element that contains the model |
---|
560 | * @throws Exception if there is some problem with extracting the |
---|
561 | * parameters. |
---|
562 | */ |
---|
563 | protected void readParameterList(Element model) throws Exception { |
---|
564 | NodeList paramL = model.getElementsByTagName("ParameterList"); |
---|
565 | |
---|
566 | // should be just one parameter list |
---|
567 | if (paramL.getLength() == 1) { |
---|
568 | Node paramN = paramL.item(0); |
---|
569 | if (paramN.getNodeType() == Node.ELEMENT_NODE) { |
---|
570 | NodeList parameterList = ((Element)paramN).getElementsByTagName("Parameter"); |
---|
571 | for (int i = 0; i < parameterList.getLength(); i++) { |
---|
572 | Node parameter = parameterList.item(i); |
---|
573 | if (parameter.getNodeType() == Node.ELEMENT_NODE) { |
---|
574 | Parameter p = new Parameter(); |
---|
575 | p.m_name = ((Element)parameter).getAttribute("name"); |
---|
576 | String label = ((Element)parameter).getAttribute("label"); |
---|
577 | if (label != null && label.length() > 0) { |
---|
578 | p.m_label = label; |
---|
579 | } |
---|
580 | m_parameterList.add(p); |
---|
581 | } |
---|
582 | } |
---|
583 | } |
---|
584 | } else { |
---|
585 | throw new Exception("[GeneralRegression] more than one parameter list!"); |
---|
586 | } |
---|
587 | } |
---|
588 | |
---|
589 | /** |
---|
590 | * Read the lists of factors and covariates. |
---|
591 | * |
---|
592 | * @param model the Element that contains the model |
---|
593 | * @param factorOrCovariate holds the String "FactorList" or |
---|
594 | * "CovariateList" |
---|
595 | * @throws Exception if there is a factor or covariate listed |
---|
596 | * that isn't in the mining schema |
---|
597 | */ |
---|
598 | protected void readFactorsAndCovariates(Element model, |
---|
599 | String factorOrCovariate) |
---|
600 | throws Exception { |
---|
601 | Instances miningSchemaI = m_miningSchema.getFieldsAsInstances(); |
---|
602 | |
---|
603 | NodeList factorL = model.getElementsByTagName(factorOrCovariate); |
---|
604 | if (factorL.getLength() == 1) { // should be 0 or 1 FactorList element |
---|
605 | Node factor = factorL.item(0); |
---|
606 | if (factor.getNodeType() == Node.ELEMENT_NODE) { |
---|
607 | NodeList predL = ((Element)factor).getElementsByTagName("Predictor"); |
---|
608 | for (int i = 0; i < predL.getLength(); i++) { |
---|
609 | Node pred = predL.item(i); |
---|
610 | if (pred.getNodeType() == Node.ELEMENT_NODE) { |
---|
611 | Predictor p = new Predictor(); |
---|
612 | p.m_name = ((Element)pred).getAttribute("name"); |
---|
613 | // find the index of this predictor in the mining schema |
---|
614 | boolean found = false; |
---|
615 | for (int j = 0; j < miningSchemaI.numAttributes(); j++) { |
---|
616 | if (miningSchemaI.attribute(j).name().equals(p.m_name)) { |
---|
617 | found = true; |
---|
618 | p.m_miningSchemaIndex = j; |
---|
619 | break; |
---|
620 | } |
---|
621 | } |
---|
622 | if (found) { |
---|
623 | if (factorOrCovariate.equals("FactorList")) { |
---|
624 | m_factorList.add(p); |
---|
625 | } else { |
---|
626 | m_covariateList.add(p); |
---|
627 | } |
---|
628 | } else { |
---|
629 | throw new Exception("[GeneralRegression] reading factors and covariates - " |
---|
630 | + "unable to find predictor " + |
---|
631 | p.m_name + " in the mining schema"); |
---|
632 | } |
---|
633 | } |
---|
634 | } |
---|
635 | } |
---|
636 | } else if (factorL.getLength() > 1){ |
---|
637 | throw new Exception("[GeneralRegression] more than one " + factorOrCovariate |
---|
638 | + "! "); |
---|
639 | } |
---|
640 | } |
---|
641 | |
---|
642 | /** |
---|
643 | * Read the PPMatrix from the xml. Does not handle multiple PPMatrixes yet. |
---|
644 | * |
---|
645 | * @param model the Element that contains the model |
---|
646 | * @throws Exception if there is a problem parsing cell values. |
---|
647 | */ |
---|
648 | protected void readPPMatrix(Element model) throws Exception { |
---|
649 | Instances miningSchemaI = m_miningSchema.getFieldsAsInstances(); |
---|
650 | |
---|
651 | NodeList matrixL = model.getElementsByTagName("PPMatrix"); |
---|
652 | |
---|
653 | // should be exactly one PPMatrix |
---|
654 | if (matrixL.getLength() == 1) { |
---|
655 | // allocate space for the matrix |
---|
656 | // column that corresponds to the class will be empty (and will be missed out |
---|
657 | // when printing the model). |
---|
658 | m_ppMatrix = new PPCell[m_parameterList.size()][miningSchemaI.numAttributes()]; |
---|
659 | |
---|
660 | Node ppM = matrixL.item(0); |
---|
661 | if (ppM.getNodeType() == Node.ELEMENT_NODE) { |
---|
662 | NodeList cellL = ((Element)ppM).getElementsByTagName("PPCell"); |
---|
663 | for (int i = 0; i < cellL.getLength(); i++) { |
---|
664 | Node cell = cellL.item(i); |
---|
665 | if (cell.getNodeType() == Node.ELEMENT_NODE) { |
---|
666 | String predictorName = ((Element)cell).getAttribute("predictorName"); |
---|
667 | String parameterName = ((Element)cell).getAttribute("parameterName"); |
---|
668 | String value = ((Element)cell).getAttribute("value"); |
---|
669 | double expOrIndex = -1; |
---|
670 | int predictorIndex = -1; |
---|
671 | int parameterIndex = -1; |
---|
672 | for (int j = 0; j < m_parameterList.size(); j++) { |
---|
673 | if (m_parameterList.get(j).m_name.equals(parameterName)) { |
---|
674 | parameterIndex = j; |
---|
675 | break; |
---|
676 | } |
---|
677 | } |
---|
678 | if (parameterIndex == -1) { |
---|
679 | throw new Exception("[GeneralRegression] unable to find parameter name " |
---|
680 | + parameterName + " in parameter list"); |
---|
681 | } |
---|
682 | |
---|
683 | Predictor p = getCovariate(predictorName); |
---|
684 | if (p != null) { |
---|
685 | try { |
---|
686 | expOrIndex = Double.parseDouble(value); |
---|
687 | predictorIndex = p.m_miningSchemaIndex; |
---|
688 | } catch (IllegalArgumentException ex) { |
---|
689 | throw new Exception("[GeneralRegression] unable to parse PPCell value: " |
---|
690 | + value); |
---|
691 | } |
---|
692 | } else { |
---|
693 | // try as a factor |
---|
694 | p = getFactor(predictorName); |
---|
695 | if (p != null) { |
---|
696 | // An example pmml file from DMG seems to suggest that it |
---|
697 | // is possible for a continuous variable in the mining schema |
---|
698 | // to be treated as a factor, so we have to check for this |
---|
699 | if (miningSchemaI.attribute(p.m_miningSchemaIndex).isNumeric()) { |
---|
700 | // parse this value as a double. It will be treated as a value |
---|
701 | // to match rather than an exponent since we are dealing with |
---|
702 | // a factor here |
---|
703 | try { |
---|
704 | expOrIndex = Double.parseDouble(value); |
---|
705 | } catch (IllegalArgumentException ex) { |
---|
706 | throw new Exception("[GeneralRegresion] unable to parse PPCell value: " |
---|
707 | + value); |
---|
708 | } |
---|
709 | } else { |
---|
710 | // it is a nominal attribute in the mining schema so find |
---|
711 | // the index that correponds to this value |
---|
712 | Attribute att = miningSchemaI.attribute(p.m_miningSchemaIndex); |
---|
713 | expOrIndex = att.indexOfValue(value); |
---|
714 | if (expOrIndex == -1) { |
---|
715 | throw new Exception("[GeneralRegression] unable to find PPCell value " |
---|
716 | + value + " in mining schema attribute " |
---|
717 | + att.name()); |
---|
718 | } |
---|
719 | } |
---|
720 | } else { |
---|
721 | throw new Exception("[GeneralRegression] cant find predictor " |
---|
722 | + predictorName + "in either the factors list " |
---|
723 | + "or the covariates list"); |
---|
724 | } |
---|
725 | predictorIndex = p.m_miningSchemaIndex; |
---|
726 | } |
---|
727 | |
---|
728 | // fill in cell value |
---|
729 | PPCell ppc = new PPCell(); |
---|
730 | ppc.m_predictorName = predictorName; ppc.m_parameterName = parameterName; |
---|
731 | ppc.m_value = expOrIndex; |
---|
732 | |
---|
733 | // TO-DO: ppc.m_targetCategory (when handling for multiple PPMatrixes is implemented) |
---|
734 | m_ppMatrix[parameterIndex][predictorIndex] = ppc; |
---|
735 | } |
---|
736 | } |
---|
737 | } |
---|
738 | } else { |
---|
739 | throw new Exception("[GeneralRegression] more than one PPMatrix!"); |
---|
740 | } |
---|
741 | } |
---|
742 | |
---|
743 | private Predictor getCovariate(String predictorName) { |
---|
744 | for (int i = 0; i < m_covariateList.size(); i++) { |
---|
745 | if (predictorName.equals(m_covariateList.get(i).m_name)) { |
---|
746 | return m_covariateList.get(i); |
---|
747 | } |
---|
748 | } |
---|
749 | return null; |
---|
750 | } |
---|
751 | |
---|
752 | private Predictor getFactor(String predictorName) { |
---|
753 | for (int i = 0; i < m_factorList.size(); i++) { |
---|
754 | if (predictorName.equals(m_factorList.get(i).m_name)) { |
---|
755 | return m_factorList.get(i); |
---|
756 | } |
---|
757 | } |
---|
758 | return null; |
---|
759 | } |
---|
760 | |
---|
761 | /** |
---|
762 | * Read the parameter matrix from the xml. |
---|
763 | * |
---|
764 | * @param model Element that holds the model |
---|
765 | * @throws Exception if a problem is encountered during extraction of |
---|
766 | * the parameter matrix |
---|
767 | */ |
---|
768 | private void readParamMatrix(Element model) throws Exception { |
---|
769 | |
---|
770 | Instances miningSchemaI = m_miningSchema.getFieldsAsInstances(); |
---|
771 | Attribute classAtt = miningSchemaI.classAttribute(); |
---|
772 | // used when function type is classification but class attribute is numeric |
---|
773 | // in the mining schema. We will assume that there is a Target specified in |
---|
774 | // the pmml that defines the legal values for this class. |
---|
775 | ArrayList<String> targetVals = null; |
---|
776 | |
---|
777 | NodeList matrixL = model.getElementsByTagName("ParamMatrix"); |
---|
778 | if (matrixL.getLength() != 1) { |
---|
779 | throw new Exception("[GeneralRegression] more than one ParamMatrix!"); |
---|
780 | } |
---|
781 | Element matrix = (Element)matrixL.item(0); |
---|
782 | |
---|
783 | |
---|
784 | // check for the case where the class in the mining schema is numeric, |
---|
785 | // but this attribute is treated as discrete |
---|
786 | if (m_functionType == Regression.RegressionTable.CLASSIFICATION && |
---|
787 | classAtt.isNumeric()) { |
---|
788 | // try and convert the class attribute to nominal. For this to succeed |
---|
789 | // there has to be a Target element defined in the PMML. |
---|
790 | if (!m_miningSchema.hasTargetMetaData()) { |
---|
791 | throw new Exception("[GeneralRegression] function type is classification and " |
---|
792 | + "class attribute in mining schema is numeric, however, " |
---|
793 | + "there is no Target element " |
---|
794 | + "specifying legal discrete values for the target!"); |
---|
795 | |
---|
796 | } |
---|
797 | |
---|
798 | if (m_miningSchema.getTargetMetaData().getOptype() |
---|
799 | != TargetMetaInfo.Optype.CATEGORICAL) { |
---|
800 | throw new Exception("[GeneralRegression] function type is classification and " |
---|
801 | + "class attribute in mining schema is numeric, however " |
---|
802 | + "Target element in PMML does not have optype categorical!"); |
---|
803 | } |
---|
804 | |
---|
805 | // OK now get legal values |
---|
806 | targetVals = m_miningSchema.getTargetMetaData().getValues(); |
---|
807 | if (targetVals.size() == 0) { |
---|
808 | throw new Exception("[GeneralRegression] function type is classification and " |
---|
809 | + "class attribute in mining schema is numeric, however " |
---|
810 | + "Target element in PMML does not have any discrete values " |
---|
811 | + "defined!"); |
---|
812 | } |
---|
813 | |
---|
814 | // Finally, convert the class in the mining schema to nominal |
---|
815 | m_miningSchema.convertNumericAttToNominal(miningSchemaI.classIndex(), targetVals); |
---|
816 | } |
---|
817 | |
---|
818 | // allocate space for the matrix |
---|
819 | m_paramMatrix = |
---|
820 | new PCell[(classAtt.isNumeric()) |
---|
821 | ? 1 |
---|
822 | : classAtt.numValues()][m_parameterList.size()]; |
---|
823 | |
---|
824 | NodeList pcellL = matrix.getElementsByTagName("PCell"); |
---|
825 | for (int i = 0; i < pcellL.getLength(); i++) { |
---|
826 | // indicates that that this beta applies to all target categories |
---|
827 | // or target is numeric |
---|
828 | int targetCategoryIndex = -1; |
---|
829 | int parameterIndex = -1; |
---|
830 | Node pcell = pcellL.item(i); |
---|
831 | if (pcell.getNodeType() == Node.ELEMENT_NODE) { |
---|
832 | String paramName = ((Element)pcell).getAttribute("parameterName"); |
---|
833 | String targetCatName = ((Element)pcell).getAttribute("targetCategory"); |
---|
834 | String coefficient = ((Element)pcell).getAttribute("beta"); |
---|
835 | String df = ((Element)pcell).getAttribute("df"); |
---|
836 | |
---|
837 | for (int j = 0; j < m_parameterList.size(); j++) { |
---|
838 | if (m_parameterList.get(j).m_name.equals(paramName)) { |
---|
839 | parameterIndex = j; |
---|
840 | // use the label if defined |
---|
841 | if (m_parameterList.get(j).m_label != null) { |
---|
842 | paramName = m_parameterList.get(j).m_label; |
---|
843 | } |
---|
844 | break; |
---|
845 | } |
---|
846 | } |
---|
847 | if (parameterIndex == -1) { |
---|
848 | throw new Exception("[GeneralRegression] unable to find parameter name " |
---|
849 | + paramName + " in parameter list"); |
---|
850 | } |
---|
851 | |
---|
852 | if (targetCatName != null && targetCatName.length() > 0) { |
---|
853 | if (classAtt.isNominal() || classAtt.isString()) { |
---|
854 | targetCategoryIndex = classAtt.indexOfValue(targetCatName); |
---|
855 | } else { |
---|
856 | throw new Exception("[GeneralRegression] found a PCell with a named " |
---|
857 | + "target category: " + targetCatName |
---|
858 | + " but class attribute is numeric in " |
---|
859 | + "mining schema"); |
---|
860 | } |
---|
861 | } |
---|
862 | |
---|
863 | PCell p = new PCell(); |
---|
864 | if (targetCategoryIndex != -1) { |
---|
865 | p.m_targetCategory = targetCatName; |
---|
866 | } |
---|
867 | p.m_parameterName = paramName; |
---|
868 | try { |
---|
869 | p.m_beta = Double.parseDouble(coefficient); |
---|
870 | } catch (IllegalArgumentException ex) { |
---|
871 | throw new Exception("[GeneralRegression] unable to parse beta value " |
---|
872 | + coefficient + " as a double from PCell"); |
---|
873 | } |
---|
874 | if (df != null && df.length() > 0) { |
---|
875 | try { |
---|
876 | p.m_df = Integer.parseInt(df); |
---|
877 | } catch (IllegalArgumentException ex) { |
---|
878 | throw new Exception("[GeneralRegression] unable to parse df value " |
---|
879 | + df + " as an int from PCell"); |
---|
880 | } |
---|
881 | } |
---|
882 | |
---|
883 | if (targetCategoryIndex != -1) { |
---|
884 | m_paramMatrix[targetCategoryIndex][parameterIndex] = p; |
---|
885 | } else { |
---|
886 | // this PCell to all target categories (covers numeric class, in |
---|
887 | // which case there will be only one row in the matrix anyway) |
---|
888 | for (int j = 0; j < m_paramMatrix.length; j++) { |
---|
889 | m_paramMatrix[j][parameterIndex] = p; |
---|
890 | } |
---|
891 | } |
---|
892 | } |
---|
893 | } |
---|
894 | } |
---|
895 | |
---|
896 | /** |
---|
897 | * Return a textual description of this general regression. |
---|
898 | * |
---|
899 | * @return a description of this general regression |
---|
900 | */ |
---|
901 | public String toString() { |
---|
902 | StringBuffer temp = new StringBuffer(); |
---|
903 | temp.append("PMML version " + getPMMLVersion()); |
---|
904 | if (!getCreatorApplication().equals("?")) { |
---|
905 | temp.append("\nApplication: " + getCreatorApplication()); |
---|
906 | } |
---|
907 | temp.append("\nPMML Model: " + m_modelType); |
---|
908 | temp.append("\n\n"); |
---|
909 | temp.append(m_miningSchema); |
---|
910 | |
---|
911 | if (m_factorList.size() > 0) { |
---|
912 | temp.append("Factors:\n"); |
---|
913 | for (Predictor p : m_factorList) { |
---|
914 | temp.append("\t" + p + "\n"); |
---|
915 | } |
---|
916 | } |
---|
917 | temp.append("\n"); |
---|
918 | if (m_covariateList.size() > 0) { |
---|
919 | temp.append("Covariates:\n"); |
---|
920 | for (Predictor p : m_covariateList) { |
---|
921 | temp.append("\t" + p + "\n"); |
---|
922 | } |
---|
923 | } |
---|
924 | temp.append("\n"); |
---|
925 | |
---|
926 | printPPMatrix(temp); |
---|
927 | temp.append("\n"); |
---|
928 | printParameterMatrix(temp); |
---|
929 | |
---|
930 | // do the link function stuff |
---|
931 | temp.append("\n"); |
---|
932 | |
---|
933 | if (m_linkFunction != LinkFunction.NONE) { |
---|
934 | temp.append("Link function: " + m_linkFunction); |
---|
935 | if (m_offsetVariable != null) { |
---|
936 | temp.append("\n\tOffset variable " + m_offsetVariable); |
---|
937 | } else if (!Double.isNaN(m_offsetValue)) { |
---|
938 | temp.append("\n\tOffset value " + m_offsetValue); |
---|
939 | } |
---|
940 | |
---|
941 | if (m_trialsVariable != null) { |
---|
942 | temp.append("\n\tTrials variable " + m_trialsVariable); |
---|
943 | } else if (!Double.isNaN(m_trialsValue)) { |
---|
944 | temp.append("\n\tTrials value " + m_trialsValue); |
---|
945 | } |
---|
946 | |
---|
947 | if (m_distribution != Distribution.NONE) { |
---|
948 | temp.append("\nDistribution: " + m_distribution); |
---|
949 | } |
---|
950 | |
---|
951 | if (m_linkFunction == LinkFunction.NEGBIN && |
---|
952 | m_distribution == Distribution.NEGBINOMIAL && |
---|
953 | !Double.isNaN(m_distParameter)) { |
---|
954 | temp.append("\n\tDistribution parameter " + m_distParameter); |
---|
955 | } |
---|
956 | |
---|
957 | if (m_linkFunction == LinkFunction.POWER || |
---|
958 | m_linkFunction == LinkFunction.ODDSPOWER) { |
---|
959 | if (!Double.isNaN(m_linkParameter)) { |
---|
960 | temp.append("\n\nLink parameter " + m_linkParameter); |
---|
961 | } |
---|
962 | } |
---|
963 | } |
---|
964 | |
---|
965 | if (m_cumulativeLinkFunction != CumulativeLinkFunction.NONE) { |
---|
966 | temp.append("Cumulative link function: " + m_cumulativeLinkFunction); |
---|
967 | |
---|
968 | if (m_offsetVariable != null) { |
---|
969 | temp.append("\n\tOffset variable " + m_offsetVariable); |
---|
970 | } else if (!Double.isNaN(m_offsetValue)) { |
---|
971 | temp.append("\n\tOffset value " + m_offsetValue); |
---|
972 | } |
---|
973 | } |
---|
974 | temp.append("\n"); |
---|
975 | |
---|
976 | return temp.toString(); |
---|
977 | } |
---|
978 | |
---|
979 | /** |
---|
980 | * Format and print the PPMatrix to the supplied StringBuffer. |
---|
981 | * |
---|
982 | * @param buff the StringBuffer to append to |
---|
983 | */ |
---|
984 | protected void printPPMatrix(StringBuffer buff) { |
---|
985 | Instances miningSchemaI = m_miningSchema.getFieldsAsInstances(); |
---|
986 | int maxAttWidth = 0; |
---|
987 | for (int i = 0; i < miningSchemaI.numAttributes(); i++) { |
---|
988 | Attribute a = miningSchemaI.attribute(i); |
---|
989 | if (a.name().length() > maxAttWidth) { |
---|
990 | maxAttWidth = a.name().length(); |
---|
991 | } |
---|
992 | } |
---|
993 | |
---|
994 | // check the width of the values |
---|
995 | for (int i = 0; i < m_parameterList.size(); i++) { |
---|
996 | for (int j = 0; j < miningSchemaI.numAttributes(); j++) { |
---|
997 | if (m_ppMatrix[i][j] != null) { |
---|
998 | double width = Math.log(Math.abs(m_ppMatrix[i][j].m_value)) / |
---|
999 | Math.log(10.0); |
---|
1000 | if (width < 0) { |
---|
1001 | width = 1; |
---|
1002 | } |
---|
1003 | // decimal + # decimal places + 1 |
---|
1004 | width += 2.0; |
---|
1005 | if ((int)width > maxAttWidth) { |
---|
1006 | maxAttWidth = (int)width; |
---|
1007 | } |
---|
1008 | if (miningSchemaI.attribute(j).isNominal() || |
---|
1009 | miningSchemaI.attribute(j).isString()) { |
---|
1010 | // check the width of this value |
---|
1011 | String val = miningSchemaI.attribute(j).value((int)m_ppMatrix[i][j].m_value) + " "; |
---|
1012 | if (val.length() > maxAttWidth) { |
---|
1013 | maxAttWidth = val.length(); |
---|
1014 | } |
---|
1015 | } |
---|
1016 | } |
---|
1017 | } |
---|
1018 | } |
---|
1019 | |
---|
1020 | // get the max parameter width |
---|
1021 | int maxParamWidth = "Parameter ".length(); |
---|
1022 | for (Parameter p : m_parameterList) { |
---|
1023 | String temp = (p.m_label != null) |
---|
1024 | ? p.m_label + " " |
---|
1025 | : p.m_name + " "; |
---|
1026 | |
---|
1027 | if (temp.length() > maxParamWidth) { |
---|
1028 | maxParamWidth = temp.length(); |
---|
1029 | } |
---|
1030 | } |
---|
1031 | |
---|
1032 | buff.append("Predictor-to-Parameter matrix:\n"); |
---|
1033 | buff.append(PMMLUtils.pad("Predictor", " ", (maxParamWidth + (maxAttWidth * 2 + 2)) |
---|
1034 | - "Predictor".length(), true)); |
---|
1035 | buff.append("\n" + PMMLUtils.pad("Parameter", " ", maxParamWidth - "Parameter".length(), false)); |
---|
1036 | // attribute names |
---|
1037 | for (int i = 0; i < miningSchemaI.numAttributes(); i++) { |
---|
1038 | if (i != miningSchemaI.classIndex()) { |
---|
1039 | String attName = miningSchemaI.attribute(i).name(); |
---|
1040 | buff.append(PMMLUtils.pad(attName, " ", maxAttWidth + 1 - attName.length(), true)); |
---|
1041 | } |
---|
1042 | } |
---|
1043 | buff.append("\n"); |
---|
1044 | |
---|
1045 | for (int i = 0; i < m_parameterList.size(); i++) { |
---|
1046 | Parameter param = m_parameterList.get(i); |
---|
1047 | String paramS = (param.m_label != null) |
---|
1048 | ? param.m_label |
---|
1049 | : param.m_name; |
---|
1050 | buff.append(PMMLUtils.pad(paramS, " ", |
---|
1051 | maxParamWidth - paramS.length(), false)); |
---|
1052 | for (int j = 0; j < miningSchemaI.numAttributes(); j++) { |
---|
1053 | if (j != miningSchemaI.classIndex()) { |
---|
1054 | PPCell p = m_ppMatrix[i][j]; |
---|
1055 | String val = " "; |
---|
1056 | if (p != null) { |
---|
1057 | if (miningSchemaI.attribute(j).isNominal() || |
---|
1058 | miningSchemaI.attribute(j).isString()) { |
---|
1059 | val = miningSchemaI.attribute(j).value((int)p.m_value); |
---|
1060 | } else { |
---|
1061 | val = "" + Utils.doubleToString(p.m_value, maxAttWidth, 4).trim(); |
---|
1062 | } |
---|
1063 | } |
---|
1064 | buff.append(PMMLUtils.pad(val, " ", maxAttWidth + 1 - val.length(), true)); |
---|
1065 | } |
---|
1066 | } |
---|
1067 | buff.append("\n"); |
---|
1068 | } |
---|
1069 | } |
---|
1070 | |
---|
1071 | /** |
---|
1072 | * Format and print the parameter matrix to the supplied StringBuffer. |
---|
1073 | * |
---|
1074 | * @param buff the StringBuffer to append to |
---|
1075 | */ |
---|
1076 | protected void printParameterMatrix(StringBuffer buff) { |
---|
1077 | Instances miningSchemaI = m_miningSchema.getFieldsAsInstances(); |
---|
1078 | |
---|
1079 | // get the maximum class value width (nominal) |
---|
1080 | int maxClassWidth = miningSchemaI.classAttribute().name().length(); |
---|
1081 | if (miningSchemaI.classAttribute().isNominal() |
---|
1082 | || miningSchemaI.classAttribute().isString()) { |
---|
1083 | for (int i = 0; i < miningSchemaI.classAttribute().numValues(); i++) { |
---|
1084 | if (miningSchemaI.classAttribute().value(i).length() > maxClassWidth) { |
---|
1085 | maxClassWidth = miningSchemaI.classAttribute().value(i).length(); |
---|
1086 | } |
---|
1087 | } |
---|
1088 | } |
---|
1089 | |
---|
1090 | // get the maximum parameter name/label width |
---|
1091 | int maxParamWidth = 0; |
---|
1092 | for (int i = 0; i < m_parameterList.size(); i++) { |
---|
1093 | Parameter p = m_parameterList.get(i); |
---|
1094 | String val = (p.m_label != null) |
---|
1095 | ? p.m_label + " " |
---|
1096 | : p.m_name + " "; |
---|
1097 | if (val.length() > maxParamWidth) { |
---|
1098 | maxParamWidth = val.length(); |
---|
1099 | } |
---|
1100 | } |
---|
1101 | |
---|
1102 | // get the max beta value width |
---|
1103 | int maxBetaWidth = "Coeff.".length(); |
---|
1104 | for (int i = 0; i < m_paramMatrix.length; i++) { |
---|
1105 | for (int j = 0; j < m_parameterList.size(); j++) { |
---|
1106 | PCell p = m_paramMatrix[i][j]; |
---|
1107 | if (p != null) { |
---|
1108 | double width = Math.log(Math.abs(p.m_beta)) / Math.log(10); |
---|
1109 | if (width < 0) { |
---|
1110 | width = 1; |
---|
1111 | } |
---|
1112 | // decimal + # decimal places + 1 |
---|
1113 | width += 7.0; |
---|
1114 | if ((int)width > maxBetaWidth) { |
---|
1115 | maxBetaWidth = (int)width; |
---|
1116 | } |
---|
1117 | } |
---|
1118 | } |
---|
1119 | } |
---|
1120 | |
---|
1121 | buff.append("Parameter estimates:\n"); |
---|
1122 | buff.append(PMMLUtils.pad(miningSchemaI.classAttribute().name(), " ", |
---|
1123 | maxClassWidth + maxParamWidth + 2 - |
---|
1124 | miningSchemaI.classAttribute().name().length(), false)); |
---|
1125 | buff.append(PMMLUtils.pad("Coeff.", " ", maxBetaWidth + 1 - "Coeff.".length(), true)); |
---|
1126 | buff.append(PMMLUtils.pad("df", " ", maxBetaWidth - "df".length(), true)); |
---|
1127 | buff.append("\n"); |
---|
1128 | for (int i = 0; i < m_paramMatrix.length; i++) { |
---|
1129 | // scan for non-null entry for this class value |
---|
1130 | boolean ok = false; |
---|
1131 | for (int j = 0; j < m_parameterList.size(); j++) { |
---|
1132 | if (m_paramMatrix[i][j] != null) { |
---|
1133 | ok = true; |
---|
1134 | } |
---|
1135 | } |
---|
1136 | if (!ok) { |
---|
1137 | continue; |
---|
1138 | } |
---|
1139 | // first the class value (if nominal) |
---|
1140 | String cVal = (miningSchemaI.classAttribute().isNominal() || |
---|
1141 | miningSchemaI.classAttribute().isString()) |
---|
1142 | ? miningSchemaI.classAttribute().value(i) |
---|
1143 | : " "; |
---|
1144 | buff.append(PMMLUtils.pad(cVal, " ", maxClassWidth - cVal.length(), false)); |
---|
1145 | buff.append("\n"); |
---|
1146 | for (int j = 0; j < m_parameterList.size(); j++) { |
---|
1147 | PCell p = m_paramMatrix[i][j]; |
---|
1148 | if (p != null) { |
---|
1149 | String label = p.m_parameterName; |
---|
1150 | buff.append(PMMLUtils.pad(label, " ", maxClassWidth + maxParamWidth + 2 - |
---|
1151 | label.length(), true)); |
---|
1152 | String betaS = Utils.doubleToString(p.m_beta, maxBetaWidth, 4).trim(); |
---|
1153 | buff.append(PMMLUtils.pad(betaS, " ", maxBetaWidth + 1 - betaS.length(), true)); |
---|
1154 | String dfS = Utils.doubleToString(p.m_df, maxBetaWidth, 4).trim(); |
---|
1155 | buff.append(PMMLUtils.pad(dfS, " ", maxBetaWidth - dfS.length(), true)); |
---|
1156 | buff.append("\n"); |
---|
1157 | } |
---|
1158 | } |
---|
1159 | } |
---|
1160 | } |
---|
1161 | |
---|
1162 | /** |
---|
1163 | * Construct the incoming parameter vector based on the values |
---|
1164 | * in the incoming test instance. |
---|
1165 | * |
---|
1166 | * @param incomingInst the values of the incoming test instance |
---|
1167 | * @return the populated parameter vector ready to be multiplied against |
---|
1168 | * the vector of coefficients. |
---|
1169 | * @throws Exception if there is some problem whilst constructing the |
---|
1170 | * parameter vector |
---|
1171 | */ |
---|
1172 | private double[] incomingParamVector(double[] incomingInst) throws Exception { |
---|
1173 | Instances miningSchemaI = m_miningSchema.getFieldsAsInstances(); |
---|
1174 | double[] incomingPV = new double[m_parameterList.size()]; |
---|
1175 | |
---|
1176 | for (int i = 0; i < m_parameterList.size(); i++) { |
---|
1177 | // |
---|
1178 | // default is that this row represents the intercept. |
---|
1179 | // this will be the case if there are all null entries in this row |
---|
1180 | incomingPV[i] = 1.0; |
---|
1181 | |
---|
1182 | // loop over the attributes (predictors) |
---|
1183 | for (int j = 0; j < miningSchemaI.numAttributes(); j++) { |
---|
1184 | PPCell cellEntry = m_ppMatrix[i][j]; |
---|
1185 | Predictor p = null; |
---|
1186 | if (cellEntry != null) { |
---|
1187 | if ((p = getFactor(cellEntry.m_predictorName)) != null) { |
---|
1188 | if ((int)incomingInst[p.m_miningSchemaIndex] == (int)cellEntry.m_value) { |
---|
1189 | incomingPV[i] *= 1.0; // we have a match |
---|
1190 | } else { |
---|
1191 | incomingPV[i] *= 0.0; |
---|
1192 | } |
---|
1193 | } else if ((p = getCovariate(cellEntry.m_predictorName)) != null) { |
---|
1194 | incomingPV[i] *= Math.pow(incomingInst[p.m_miningSchemaIndex], cellEntry.m_value); |
---|
1195 | } else { |
---|
1196 | throw new Exception("[GeneralRegression] can't find predictor " |
---|
1197 | + cellEntry.m_predictorName + " in either the list of factors or covariates"); |
---|
1198 | } |
---|
1199 | } |
---|
1200 | } |
---|
1201 | } |
---|
1202 | |
---|
1203 | return incomingPV; |
---|
1204 | } |
---|
1205 | |
---|
1206 | /** |
---|
1207 | * Classifies the given test instance. The instance has to belong to a |
---|
1208 | * dataset when it's being classified. |
---|
1209 | * |
---|
1210 | * @param inst the instance to be classified |
---|
1211 | * @return the predicted most likely class for the instance or |
---|
1212 | * Utils.missingValue() if no prediction is made |
---|
1213 | * @exception Exception if an error occurred during the prediction |
---|
1214 | */ |
---|
1215 | public double[] distributionForInstance(Instance inst) throws Exception { |
---|
1216 | if (!m_initialized) { |
---|
1217 | mapToMiningSchema(inst.dataset()); |
---|
1218 | } |
---|
1219 | double[] preds = null; |
---|
1220 | if (m_miningSchema.getFieldsAsInstances().classAttribute().isNumeric()) { |
---|
1221 | preds = new double[1]; |
---|
1222 | } else { |
---|
1223 | preds = new double[m_miningSchema.getFieldsAsInstances().classAttribute().numValues()]; |
---|
1224 | } |
---|
1225 | |
---|
1226 | // create an array of doubles that holds values from the incoming |
---|
1227 | // instance; in order of the fields in the mining schema. We will |
---|
1228 | // also handle missing values and outliers here. |
---|
1229 | double[] incoming = m_fieldsMap.instanceToSchema(inst, m_miningSchema); |
---|
1230 | |
---|
1231 | // In this implementation we will default to information in the Target element (default |
---|
1232 | // value for numeric prediction and prior probabilities for classification). If there is |
---|
1233 | // no Target element defined, then an Exception is thrown. |
---|
1234 | |
---|
1235 | boolean hasMissing = false; |
---|
1236 | for (int i = 0; i < incoming.length; i++) { |
---|
1237 | if (i != m_miningSchema.getFieldsAsInstances().classIndex() && |
---|
1238 | Double.isNaN(incoming[i])) { |
---|
1239 | hasMissing = true; |
---|
1240 | break; |
---|
1241 | } |
---|
1242 | } |
---|
1243 | |
---|
1244 | if (hasMissing) { |
---|
1245 | if (!m_miningSchema.hasTargetMetaData()) { |
---|
1246 | String message = "[GeneralRegression] WARNING: Instance to predict has missing value(s) but " |
---|
1247 | + "there is no missing value handling meta data and no " |
---|
1248 | + "prior probabilities/default value to fall back to. No " |
---|
1249 | + "prediction will be made (" |
---|
1250 | + ((m_miningSchema.getFieldsAsInstances().classAttribute().isNominal() |
---|
1251 | || m_miningSchema.getFieldsAsInstances().classAttribute().isString()) |
---|
1252 | ? "zero probabilities output)." |
---|
1253 | : "NaN output)."); |
---|
1254 | if (m_log == null) { |
---|
1255 | System.err.println(message); |
---|
1256 | } else { |
---|
1257 | m_log.logMessage(message); |
---|
1258 | } |
---|
1259 | |
---|
1260 | if (m_miningSchema.getFieldsAsInstances().classAttribute().isNumeric()) { |
---|
1261 | preds[0] = Utils.missingValue(); |
---|
1262 | } |
---|
1263 | return preds; |
---|
1264 | } else { |
---|
1265 | // use prior probablilities/default value |
---|
1266 | TargetMetaInfo targetData = m_miningSchema.getTargetMetaData(); |
---|
1267 | if (m_miningSchema.getFieldsAsInstances().classAttribute().isNumeric()) { |
---|
1268 | preds[0] = targetData.getDefaultValue(); |
---|
1269 | } else { |
---|
1270 | Instances miningSchemaI = m_miningSchema.getFieldsAsInstances(); |
---|
1271 | for (int i = 0; i < miningSchemaI.classAttribute().numValues(); i++) { |
---|
1272 | preds[i] = targetData.getPriorProbability(miningSchemaI.classAttribute().value(i)); |
---|
1273 | } |
---|
1274 | } |
---|
1275 | return preds; |
---|
1276 | } |
---|
1277 | } else { |
---|
1278 | // construct input parameter vector here |
---|
1279 | double[] inputParamVector = incomingParamVector(incoming); |
---|
1280 | computeResponses(incoming, inputParamVector, preds); |
---|
1281 | } |
---|
1282 | |
---|
1283 | return preds; |
---|
1284 | } |
---|
1285 | |
---|
1286 | /** |
---|
1287 | * Compute the responses for the function given the parameter values corresponding |
---|
1288 | * to the current incoming instance. |
---|
1289 | * |
---|
1290 | * @param incomingInst raw incoming instance values (after missing value |
---|
1291 | * replacement and outlier treatment) |
---|
1292 | * @param incomingParamVector incoming instance values mapped to parameters |
---|
1293 | * @param responses will contain the responses computed by the function |
---|
1294 | * @throws Exception if something goes wrong |
---|
1295 | */ |
---|
1296 | private void computeResponses(double[] incomingInst, |
---|
1297 | double[] incomingParamVector, |
---|
1298 | double[] responses) throws Exception { |
---|
1299 | for (int i = 0; i < responses.length; i++) { |
---|
1300 | for (int j = 0; j < m_parameterList.size(); j++) { |
---|
1301 | // a row of the parameter matrix should have all non-null entries |
---|
1302 | // except for the last class (in the case of classification) which |
---|
1303 | // should have just an intercept of 0. Need to handle the case where |
---|
1304 | // no intercept has been defined in the pmml file for the last class |
---|
1305 | PCell p = m_paramMatrix[i][j]; |
---|
1306 | if (p == null) { |
---|
1307 | responses[i] += 0.0 * incomingParamVector[j]; |
---|
1308 | } else { |
---|
1309 | responses[i] += incomingParamVector[j] * p.m_beta; |
---|
1310 | } |
---|
1311 | } |
---|
1312 | } |
---|
1313 | |
---|
1314 | switch(m_modelType) { |
---|
1315 | case MULTINOMIALLOGISTIC: |
---|
1316 | computeProbabilitiesMultinomialLogistic(responses); |
---|
1317 | break; |
---|
1318 | case REGRESSION: |
---|
1319 | // nothing to be done |
---|
1320 | break; |
---|
1321 | case GENERALLINEAR: |
---|
1322 | case GENERALIZEDLINEAR: |
---|
1323 | if (m_linkFunction != LinkFunction.NONE) { |
---|
1324 | computeResponseGeneralizedLinear(incomingInst, responses); |
---|
1325 | } else { |
---|
1326 | throw new Exception("[GeneralRegression] no link function specified!"); |
---|
1327 | } |
---|
1328 | break; |
---|
1329 | case ORDINALMULTINOMIAL: |
---|
1330 | if (m_cumulativeLinkFunction != CumulativeLinkFunction.NONE) { |
---|
1331 | computeResponseOrdinalMultinomial(incomingInst, responses); |
---|
1332 | } else { |
---|
1333 | throw new Exception("[GeneralRegression] no cumulative link function specified!"); |
---|
1334 | } |
---|
1335 | break; |
---|
1336 | default: |
---|
1337 | throw new Exception("[GeneralRegression] unknown model type"); |
---|
1338 | } |
---|
1339 | } |
---|
1340 | |
---|
1341 | /** |
---|
1342 | * Computes probabilities for the multinomial logistic model type. |
---|
1343 | * |
---|
1344 | * @param responses will hold the responses computed by the function. |
---|
1345 | */ |
---|
1346 | private static void computeProbabilitiesMultinomialLogistic(double[] responses) { |
---|
1347 | double[] r = responses.clone(); |
---|
1348 | for (int j = 0; j < r.length; j++) { |
---|
1349 | double sum = 0; |
---|
1350 | boolean overflow = false; |
---|
1351 | for (int k = 0; k < r.length; k++) { |
---|
1352 | if (r[k] - r[j] > 700) { |
---|
1353 | overflow = true; |
---|
1354 | break; |
---|
1355 | } |
---|
1356 | sum += Math.exp(r[k] - r[j]); |
---|
1357 | } |
---|
1358 | if (overflow) { |
---|
1359 | responses[j] = 0.0; |
---|
1360 | } else { |
---|
1361 | responses[j] = 1.0 / sum; |
---|
1362 | } |
---|
1363 | } |
---|
1364 | } |
---|
1365 | |
---|
1366 | /** |
---|
1367 | * Computes responses for the general linear and generalized linear model |
---|
1368 | * types. |
---|
1369 | * |
---|
1370 | * @param incomingInst the raw incoming instance values (after missing value |
---|
1371 | * replacement and outlier treatment etc). |
---|
1372 | * @param responses will hold the responses computed by the function |
---|
1373 | * @throws Exception if a problem occurs. |
---|
1374 | */ |
---|
1375 | private void computeResponseGeneralizedLinear(double[] incomingInst, |
---|
1376 | double[] responses) |
---|
1377 | throws Exception { |
---|
1378 | double[] r = responses.clone(); |
---|
1379 | |
---|
1380 | double offset = 0; |
---|
1381 | if (m_offsetVariable != null) { |
---|
1382 | Attribute offsetAtt = |
---|
1383 | m_miningSchema.getFieldsAsInstances().attribute(m_offsetVariable); |
---|
1384 | if (offsetAtt == null) { |
---|
1385 | throw new Exception("[GeneralRegression] unable to find offset variable " |
---|
1386 | + m_offsetVariable + " in the mining schema!"); |
---|
1387 | } |
---|
1388 | offset = incomingInst[offsetAtt.index()]; |
---|
1389 | } else if (!Double.isNaN(m_offsetValue)) { |
---|
1390 | offset = m_offsetValue; |
---|
1391 | } |
---|
1392 | |
---|
1393 | double trials = 1; |
---|
1394 | if (m_trialsVariable != null) { |
---|
1395 | Attribute trialsAtt = m_miningSchema.getFieldsAsInstances().attribute(m_trialsVariable); |
---|
1396 | if (trialsAtt == null) { |
---|
1397 | throw new Exception("[GeneralRegression] unable to find trials variable " |
---|
1398 | + m_trialsVariable + " in the mining schema!"); |
---|
1399 | } |
---|
1400 | trials = incomingInst[trialsAtt.index()]; |
---|
1401 | } else if (!Double.isNaN(m_trialsValue)) { |
---|
1402 | trials = m_trialsValue; |
---|
1403 | } |
---|
1404 | |
---|
1405 | double distParam = 0; |
---|
1406 | if (m_linkFunction == LinkFunction.NEGBIN && |
---|
1407 | m_distribution == Distribution.NEGBINOMIAL) { |
---|
1408 | if (Double.isNaN(m_distParameter)) { |
---|
1409 | throw new Exception("[GeneralRegression] no distribution parameter defined!"); |
---|
1410 | } |
---|
1411 | distParam = m_distParameter; |
---|
1412 | } |
---|
1413 | |
---|
1414 | double linkParam = 0; |
---|
1415 | if (m_linkFunction == LinkFunction.POWER || |
---|
1416 | m_linkFunction == LinkFunction.ODDSPOWER) { |
---|
1417 | if (Double.isNaN(m_linkParameter)) { |
---|
1418 | throw new Exception("[GeneralRegression] no link parameter defined!"); |
---|
1419 | } |
---|
1420 | linkParam = m_linkParameter; |
---|
1421 | } |
---|
1422 | |
---|
1423 | for (int i = 0; i < r.length; i++) { |
---|
1424 | responses[i] = m_linkFunction.eval(r[i], offset, trials, distParam, linkParam); |
---|
1425 | } |
---|
1426 | } |
---|
1427 | |
---|
1428 | /** |
---|
1429 | * Computes responses for the ordinal multinomial model type. |
---|
1430 | * |
---|
1431 | * @param incomingInst the raw incoming instance values (after missing value |
---|
1432 | * replacement and outlier treatment etc). |
---|
1433 | * @param responses will hold the responses computed by the function |
---|
1434 | * @throws Exception if a problem occurs. |
---|
1435 | */ |
---|
1436 | private void computeResponseOrdinalMultinomial(double[] incomingInst, |
---|
1437 | double[] responses) throws Exception { |
---|
1438 | |
---|
1439 | double[] r = responses.clone(); |
---|
1440 | |
---|
1441 | double offset = 0; |
---|
1442 | if (m_offsetVariable != null) { |
---|
1443 | Attribute offsetAtt = |
---|
1444 | m_miningSchema.getFieldsAsInstances().attribute(m_offsetVariable); |
---|
1445 | if (offsetAtt == null) { |
---|
1446 | throw new Exception("[GeneralRegression] unable to find offset variable " |
---|
1447 | + m_offsetVariable + " in the mining schema!"); |
---|
1448 | } |
---|
1449 | offset = incomingInst[offsetAtt.index()]; |
---|
1450 | } else if (!Double.isNaN(m_offsetValue)) { |
---|
1451 | offset = m_offsetValue; |
---|
1452 | } |
---|
1453 | |
---|
1454 | for (int i = 0; i < r.length; i++) { |
---|
1455 | if (i == 0) { |
---|
1456 | responses[i] = m_cumulativeLinkFunction.eval(r[i], offset); |
---|
1457 | |
---|
1458 | } else if (i == (r.length - 1)) { |
---|
1459 | responses[i] = 1.0 - responses[i - 1]; |
---|
1460 | } else { |
---|
1461 | responses[i] = m_cumulativeLinkFunction.eval(r[i], offset) - responses[i - 1]; |
---|
1462 | } |
---|
1463 | } |
---|
1464 | } |
---|
1465 | |
---|
1466 | /* (non-Javadoc) |
---|
1467 | * @see weka.core.RevisionHandler#getRevision() |
---|
1468 | */ |
---|
1469 | public String getRevision() { |
---|
1470 | return RevisionUtils.extract("$Revision: 5987 $"); |
---|
1471 | } |
---|
1472 | } |
---|