Debian Codegraph @

data/glass/glassc.influ-o-text-1

raw
1
2
3
4 I N F L U E N C E V A L U E S R E P O R T
5 order attributes by influence values = true
6 =============================================
7
8 AutoClass CLASSIFICATION for the 214 cases in
9 /home/wtaylor/AC/3.3.4/autoclass-c/data/glass/glassc.db2
10 /home/wtaylor/AC/3.3.4/autoclass-c/data/glass/glass-3c.hd2
11 with log-A<X/H> (approximate marginal likelihood) = -10897.738
12 from classification results file
13 /home/wtaylor/AC/3.3.4/autoclass-c/data/glass/glassc.results-bin
14 and using models
15 /home/wtaylor/AC/3.3.4/autoclass-c/data/glass/glass-mnc.model - index = 0
16
17
18
19
20 ORDER OF PRESENTATION:
21
22 * Summary of the generating search.
23 * Weight ordered list of the classes found & class strength heuristic.
24 * List of class cross entropies with respect to the global class.
25 * Ordered list of attribute influence values summed over all classes.
26 * Class listings, ordered by class weight.
27
28
29
30
31
32
33
34 _____________________________________________________________________________
35 _____________________________________________________________________________
36
37 SEARCH SUMMARY 4 tries over 2 seconds
38
39 _______________ SUMMARY OF 10 BEST RESULTS _________________________ ##
40 PROBABILITY exp(-10897.738) N_CLASSES 5 FOUND ON TRY 3 *SAVED* -1
41 PROBABILITY exp(-11187.745) N_CLASSES 7 FOUND ON TRY 4 *SAVED* -2
42 PROBABILITY exp(-11229.516) N_CLASSES 3 FOUND ON TRY 2
43 PROBABILITY exp(-11236.432) N_CLASSES 2 FOUND ON TRY 1
44
45
46 ## - report filenames suffix
47 _____________________________________________________________________________
48 _____________________________________________________________________________
49
50
51 CLASSIFICATION HAS 5 POPULATED CLASSES (max global influence value = 1.917)
52
53 We give below a heuristic measure of class strength: the approximate
54 geometric mean probability for instances belonging to each class,
55 computed from the class parameters and statistics. This approximates
56 the contribution made, by any one instance "belonging" to the class,
57 to the log probability of the data set w.r.t. the classification. It
58 thus provides a heuristic measure of how strongly each class predicts
59 "its" instances.
60
61 Class Log of class Relative Class Normalized
62 num strength class strength weight class weight
63
64 0 -4.21e+01 1.00e+00 97 0.453
65 1 -5.03e+01 2.77e-04 46 0.215
66 2 -5.82e+01 1.01e-07 35 0.163
67 3 -5.54e+01 1.62e-06 19 0.089
68 4 -5.39e+01 6.97e-06 17 0.080
69
70
71 CLASS DIVERGENCES
72
73 The class divergence, or cross entropy w.r.t. the single class
74 classification, is a measure of how strongly the class probability
75 distribution function differs from that of the database as a whole.
76 It is zero for identical distributions, going infinite when two
77 discrete distributions place probability 1 on differing values of the
78 same attribute.
79
80 Class class cross entropy Class Normalized
81 num w.r.t. global class weight class weight
82
83 0 1.31e+01 97 0.453
84 1 8.16e+00 46 0.215
85 2 5.68e+00 35 0.163
86 3 6.25e+00 19 0.089
87 4 1.73e+01 17 0.080
88
89 ORDERED LIST OF NORMALIZED ATTRIBUTE INFLUENCE VALUES SUMMED OVER ALL CLASSES
90
91 This gives a rough heuristic measure of relative influence of each
92 attribute in differentiating the classes from the overall data set.
93 Note that "influence values" are only computable with respect to the
94 model terms. When multiple attributes are modeled by a single
95 dependent term (e.g. multi_normal_cn), the term influence value is
96 distributed equally over the modeled attributes.
97
98 num description I-*k
99
100 011 Log RI: refractive index 1.000
101 012 Log Na: Wt.% Sodium oxide 1.000
102 013 Log Mg: Wt.% Magnesium oxide 1.000
103 014 Log Al: Wt.% Aluminum oxide 1.000
104 015 Log Si: Wt.% Silicon oxide 1.000
105 016 Log K: Wt.% Potassium oxide 1.000
106 017 Log Ca: Wt.% Calcium oxide 1.000
107 018 Log Ba: Wt.% Barium oxide 1.000
108 019 Log Fe: Wt.% Iron oxide 1.000
109 000 Id number -----
110 001 RI: refractive index -----
111 002 Na: Wt.% Sodium oxide -----
112 003 Mg: Wt.% Magnesium oxide -----
113 004 Al: Wt.% Aluminum oxide -----
114 005 Si: Wt.% Silicon oxide -----
115 006 K: Wt.% Potassium oxide -----
116 007 Ca: Wt.% Calcium oxide -----
117 008 Ba: Wt.% Barium oxide -----
118 009 Fe: Wt.% Iron oxide -----
119 010 Type of glass -----
120
121 CLASS LISTINGS:
122
123 These listings are ordered by class weight --
124 * j is the zero-based class index,
125 * k is the zero-based attribute index, and
126 * l is the zero-based discrete attribute instance index.
127
128 Within each class, the covariant and independent model terms are ordered
129 by their term influence value I-jk.
130
131 Covariant attributes and discrete attribute instance values are both
132 ordered by their significance value. Significance values are computed
133 with respect to a single class classification, using the divergence from
134 it, abs( log( Prob-jkl / Prob-*kl)), for discrete attributes and the
135 relative separation from it, abs( Mean-jk - Mean-*k) / StDev-jk, for
136 numerical valued attributes. For the SNcm model, the value line is
137 followed by the probabilities that the value is known, for that class
138 and for the single class classification.
139
140 Entries are attribute type dependent, and the corresponding headers are
141 reproduced with each class. In these --
142 * num/t denotes model term number,
143 * num/a denotes attribute number,
144 * t denotes attribute type,
145 * mtt denotes model term type, and
146 * I-jk denotes the term influence value for attribute k
147 in class j. This is the cross entropy or
148 Kullback-Leibler distance between the class and
149 full database probability distributions (see
150 interpretation-c.text).
151 * Mean StDev
152 -jk -jk The estimated mean and standard deviation
153 for attribute k in class j.
154 * |Mean-jk - The absolute difference between the
155 Mean-*k|/ two means, scaled w.r.t. the class
156 StDev-jk standard deviation, to get a measure
157 of the distance between the attribute
158 means in the class and full data.
159 * Mean StDev The estimated mean and standard
160 -*k -*k deviation for attribute k when the
161 model is applied to the data set
162 as a whole.
163 * Prob-jk is known 1.00e+00 Prob-*k is known 9.98e-01
164 The SNcm model allows for the possibility that data
165 values are unknown, and models this with a discrete
166 known/unknown probability. The gaussian normal for
167 known values is then conditional on the known
168 probability. In this instance, we have a class
169 where all values are known, as opposed to a database
170 where only 99.8% of values are known.
171
172
173 CLASS 0 - weight 97 normalized weight 0.453 relative strength 1.00e+00 *******
174 class cross entropy w.r.t. global class 1.31e+01 *******
175
176 Model file: /home/wtaylor/AC/3.3.4/autoclass-c/data/glass/glass-mnc.model - index = 0
177 Numbers: numb/t = model term number; numb/a = attribute number
178 Model term types (mtt): (multi_normal_cn MNcn)
179
180
181 REAL ATTRIBUTE (t = R) |Mean-jk -
182 numb t mtt description I-jk Mean StDev Mean-*k|/ Mean StDev
183 t a -jk -jk StDev-jk -*k -*k
184
185 00 13 R MNcn Log Mg: Wt.% Magnesi 1.454 ( 1.27e+00 4.25e-02) 3.21e+01 (-9.37e-02 2.60e+00)
186 um oxide
187 00 18 R MNcn Log Ba: Wt.% Barium 1.454 (-5.30e+00 2.86e-02) 3.03e+01 (-4.43e+00 1.93e+00)
188 oxide
189 00 16 R MNcn Log K: Wt.% Potassi 1.454 (-5.10e-01 8.42e-02) 1.14e+01 (-1.47e+00 1.74e+00)
190 um oxide
191 00 17 R MNcn Log Ca: Wt.% Calcium 1.454 ( 2.12e+00 3.88e-02) 1.48e+00 ( 2.18e+00 1.45e-01)
192 oxide
193 00 11 R MNcn Log RI: refractive i 1.454 (-6.59e-01 1.77e-03) 1.12e+00 (-6.57e-01 5.82e-03)
194 ndex
195 00 12 R MNcn Log Na: Wt.% Sodium 1.454 ( 2.57e+00 2.50e-02) 8.66e-01 ( 2.59e+00 6.09e-02)
196 oxide
197 00 15 R MNcn Log Si: Wt.% Silicon 1.454 ( 4.29e+00 5.21e-03) 3.25e-01 ( 4.29e+00 1.07e-02)
198 oxide
199 00 14 R MNcn Log Al: Wt.% Aluminu 1.454 ( 3.18e-01 1.41e-01) 8.24e-02 ( 3.06e-01 3.70e-01)
200 m oxide
201 00 19 R MNcn Log Fe: Wt.% Iron ox 1.454 (-4.12e+00 1.69e+00) 3.34e-02 (-4.18e+00 1.66e+00)
202 ide
203
204 Correlation matrix (row & column indices are attribute numbers)
205 11 12 13 14 15 16 17 18 19
206 11 1.000 0.176 0.276 -0.445 -0.434 -0.441 0.582 0.000 0.114
207 12 0.176 1.000 0.239 0.025 -0.710 -0.326 -0.232 0.000 -0.155
208 13 0.276 0.239 1.000 -0.139 -0.266 -0.172 -0.287 0.000 0.060
209 14 -0.445 0.025 -0.139 1.000 -0.139 0.364 -0.497 0.000 -0.005
210 15 -0.434 -0.710 -0.266 -0.139 1.000 0.249 -0.207 0.000 -0.084
211 16 -0.441 -0.326 -0.172 0.364 0.249 1.000 -0.300 0.000 0.033
212 17 0.582 -0.232 -0.287 -0.497 -0.207 -0.300 1.000 0.000 0.177
213 18 0.000 0.000 0.000 0.000 0.000 0.000 0.000 1.000 0.000
214 19 0.114 -0.155 0.060 -0.005 -0.084 0.033 0.177 0.000 1.000
215
216
217
218 CLASS 1 - weight 46 normalized weight 0.215 relative strength 2.77e-04 *******
219 class cross entropy w.r.t. global class 8.16e+00 *******
220
221 Model file: /home/wtaylor/AC/3.3.4/autoclass-c/data/glass/glass-mnc.model - index = 0
222 Numbers: numb/t = model term number; numb/a = attribute number
223 Model term types (mtt): (multi_normal_cn MNcn)
224
225
226 REAL ATTRIBUTE (t = R) |Mean-jk -
227 numb t mtt description I-jk Mean StDev Mean-*k|/ Mean StDev
228 t a -jk -jk StDev-jk -*k -*k
229
230 00 18 R MNcn Log Ba: Wt.% Barium 0.906 (-5.30e+00 2.91e-02) 2.97e+01 (-4.43e+00 1.93e+00)
231 oxide
232 00 13 R MNcn Log Mg: Wt.% Magnesi 0.906 ( 1.25e+00 1.18e-01) 1.14e+01 (-9.37e-02 2.60e+00)
233 um oxide
234 00 14 R MNcn Log Al: Wt.% Aluminu 0.906 (-4.90e-05 4.09e-01) 7.49e-01 ( 3.06e-01 3.70e-01)
235 m oxide
236 00 15 R MNcn Log Si: Wt.% Silicon 0.906 ( 4.28e+00 1.01e-02) 4.38e-01 ( 4.29e+00 1.07e-02)
237 oxide
238 00 11 R MNcn Log RI: refractive i 0.906 (-6.55e-01 6.40e-03) 3.95e-01 (-6.57e-01 5.82e-03)
239 ndex
240 00 17 R MNcn Log Ca: Wt.% Calcium 0.906 ( 2.20e+00 9.66e-02) 2.28e-01 ( 2.18e+00 1.45e-01)
241 oxide
242 00 12 R MNcn Log Na: Wt.% Sodium 0.906 ( 2.60e+00 4.55e-02) 2.15e-01 ( 2.59e+00 6.09e-02)
243 oxide
244 00 16 R MNcn Log K: Wt.% Potassi 0.906 (-1.71e+00 1.20e+00) 2.01e-01 (-1.47e+00 1.74e+00)
245 um oxide
246 00 19 R MNcn Log Fe: Wt.% Iron ox 0.906 (-4.11e+00 1.70e+00) 3.70e-02 (-4.18e+00 1.66e+00)
247 ide
248
249 Correlation matrix (row & column indices are attribute numbers)
250 11 12 13 14 15 16 17 18 19
251 11 1.000 0.308 0.301 -0.605 -0.747 -0.503 0.816 0.000 -0.029
252 12 0.308 1.000 0.502 -0.413 -0.661 -0.551 0.003 0.000 -0.159
253 13 0.301 0.502 1.000 -0.426 -0.484 -0.521 0.041 0.000 -0.063
254 14 -0.605 -0.413 -0.426 1.000 0.496 0.547 -0.621 0.000 -0.013
255 15 -0.747 -0.661 -0.484 0.496 1.000 0.584 -0.556 0.000 0.122
256 16 -0.503 -0.551 -0.521 0.547 0.584 1.000 -0.430 0.000 0.103
257 17 0.816 0.003 0.041 -0.621 -0.556 -0.430 1.000 0.000 0.067
258 18 0.000 0.000 0.000 0.000 0.000 0.000 0.000 1.000 0.000
259 19 -0.029 -0.159 -0.063 -0.013 0.122 0.103 0.067 0.000 1.000
260
261
262
263 CLASS 2 - weight 35 normalized weight 0.163 relative strength 1.01e-07 *******
264 class cross entropy w.r.t. global class 5.68e+00 *******
265
266 Model file: /home/wtaylor/AC/3.3.4/autoclass-c/data/glass/glass-mnc.model - index = 0
267 Numbers: numb/t = model term number; numb/a = attribute number
268 Model term types (mtt): (multi_normal_cn MNcn)
269
270
271 REAL ATTRIBUTE (t = R) |Mean-jk -
272 numb t mtt description I-jk Mean StDev Mean-*k|/ Mean StDev
273 t a -jk -jk StDev-jk -*k -*k
274
275 00 14 R MNcn Log Al: Wt.% Aluminu 0.631 ( 7.59e-01 2.37e-01) 1.91e+00 ( 3.06e-01 3.70e-01)
276 m oxide
277 00 12 R MNcn Log Na: Wt.% Sodium 0.631 ( 2.66e+00 3.78e-02) 1.85e+00 ( 2.59e+00 6.09e-02)
278 oxide
279 00 18 R MNcn Log Ba: Wt.% Barium 0.631 (-1.34e+00 2.48e+00) 1.25e+00 (-4.43e+00 1.93e+00)
280 oxide
281 00 13 R MNcn Log Mg: Wt.% Magnesi 0.631 (-3.57e+00 2.87e+00) 1.21e+00 (-9.37e-02 2.60e+00)
282 um oxide
283 00 16 R MNcn Log K: Wt.% Potassi 0.631 (-3.78e+00 2.47e+00) 9.34e-01 (-1.47e+00 1.74e+00)
284 um oxide
285 00 11 R MNcn Log RI: refractive i 0.631 (-6.61e-01 3.93e-03) 9.02e-01 (-6.57e-01 5.82e-03)
286 ndex
287 00 19 R MNcn Log Fe: Wt.% Iron ox 0.631 (-4.87e+00 1.02e+00) 6.84e-01 (-4.18e+00 1.66e+00)
288 ide
289 00 17 R MNcn Log Ca: Wt.% Calcium 0.631 ( 2.12e+00 1.61e-01) 3.53e-01 ( 2.18e+00 1.45e-01)
290 oxide
291 00 15 R MNcn Log Si: Wt.% Silicon 0.631 ( 4.29e+00 1.33e-02) 1.76e-01 ( 4.29e+00 1.07e-02)
292 oxide
293
294 Correlation matrix (row & column indices are attribute numbers)
295 11 12 13 14 15 16 17 18 19
296 11 1.000 0.301 0.075 -0.427 0.098 -0.466 0.481 -0.078 -0.047
297 12 0.301 1.000 -0.252 -0.191 0.548 -0.625 0.301 0.402 0.088
298 13 0.075 -0.252 1.000 -0.469 -0.281 0.084 -0.254 -0.423 -0.268
299 14 -0.427 -0.191 -0.469 1.000 -0.387 0.502 -0.213 0.299 0.181
300 15 0.098 0.548 -0.281 -0.387 1.000 -0.677 0.469 0.235 0.213
301 16 -0.466 -0.625 0.084 0.502 -0.677 1.000 -0.721 0.017 -0.068
302 17 0.481 0.301 -0.254 -0.213 0.469 -0.721 1.000 -0.268 0.145
303 18 -0.078 0.402 -0.423 0.299 0.235 0.017 -0.268 1.000 0.206
304 19 -0.047 0.088 -0.268 0.181 0.213 -0.068 0.145 0.206 1.000
305
306
307
308 CLASS 3 - weight 19 normalized weight 0.089 relative strength 1.62e-06 *******
309 class cross entropy w.r.t. global class 6.25e+00 *******
310
311 Model file: /home/wtaylor/AC/3.3.4/autoclass-c/data/glass/glass-mnc.model - index = 0
312 Numbers: numb/t = model term number; numb/a = attribute number
313 Model term types (mtt): (multi_normal_cn MNcn)
314
315
316 REAL ATTRIBUTE (t = R) |Mean-jk -
317 numb t mtt description I-jk Mean StDev Mean-*k|/ Mean StDev
318 t a -jk -jk StDev-jk -*k -*k
319
320 00 13 R MNcn Log Mg: Wt.% Magnesi 0.695 ( 7.47e-01 6.16e-01) 1.37e+00 (-9.37e-02 2.60e+00)
321 um oxide
322 00 16 R MNcn Log K: Wt.% Potassi 0.695 (-8.34e-01 5.87e-01) 1.09e+00 (-1.47e+00 1.74e+00)
323 um oxide
324 00 11 R MNcn Log RI: refractive i 0.695 (-6.54e-01 3.92e-03) 7.91e-01 (-6.57e-01 5.82e-03)
325 ndex
326 00 17 R MNcn Log Ca: Wt.% Calcium 0.695 ( 2.26e+00 1.20e-01) 6.30e-01 ( 2.18e+00 1.45e-01)
327 oxide
328 00 18 R MNcn Log Ba: Wt.% Barium 0.695 (-3.18e+00 2.10e+00) 5.95e-01 (-4.43e+00 1.93e+00)
329 oxide
330 00 19 R MNcn Log Fe: Wt.% Iron ox 0.695 (-3.28e+00 2.09e+00) 4.28e-01 (-4.18e+00 1.66e+00)
331 ide
332 00 15 R MNcn Log Si: Wt.% Silicon 0.695 ( 4.28e+00 1.31e-02) 3.23e-01 ( 4.29e+00 1.07e-02)
333 oxide
334 00 14 R MNcn Log Al: Wt.% Aluminu 0.695 ( 3.67e-01 2.47e-01) 2.49e-01 ( 3.06e-01 3.70e-01)
335 m oxide
336 00 12 R MNcn Log Na: Wt.% Sodium 0.695 ( 2.59e+00 8.41e-02) 1.06e-01 ( 2.59e+00 6.09e-02)
337 oxide
338
339 Correlation matrix (row & column indices are attribute numbers)
340 11 12 13 14 15 16 17 18 19
341 11 1.000 0.278 -0.505 0.262 -0.548 -0.213 0.551 -0.215 -0.390
342 12 0.278 1.000 0.054 -0.107 -0.792 -0.220 -0.387 0.342 -0.049
343 13 -0.505 0.054 1.000 -0.354 -0.074 0.306 -0.718 0.482 -0.047
344 14 0.262 -0.107 -0.354 1.000 -0.204 0.465 0.393 -0.316 -0.145
345 15 -0.548 -0.792 -0.074 -0.204 1.000 0.013 0.148 -0.190 0.329
346 16 -0.213 -0.220 0.306 0.465 0.013 1.000 -0.091 -0.131 0.008
347 17 0.551 -0.387 -0.718 0.393 0.148 -0.091 1.000 -0.738 -0.199
348 18 -0.215 0.342 0.482 -0.316 -0.190 -0.131 -0.738 1.000 0.110
349 19 -0.390 -0.049 -0.047 -0.145 0.329 0.008 -0.199 0.110 1.000
350
351
352
353 CLASS 4 - weight 17 normalized weight 0.080 relative strength 6.97e-06 *******
354 class cross entropy w.r.t. global class 1.73e+01 *******
355
356 Model file: /home/wtaylor/AC/3.3.4/autoclass-c/data/glass/glass-mnc.model - index = 0
357 Numbers: numb/t = model term number; numb/a = attribute number
358 Model term types (mtt): (multi_normal_cn MNcn)
359
360
361 REAL ATTRIBUTE (t = R) |Mean-jk -
362 numb t mtt description I-jk Mean StDev Mean-*k|/ Mean StDev
363 t a -jk -jk StDev-jk -*k -*k
364
365 00 13 R MNcn Log Mg: Wt.% Magnesi 1.917 (-5.30e+00 2.04e-03) 2.55e+03 (-9.37e-02 2.60e+00)
366 um oxide
367 00 17 R MNcn Log Ca: Wt.% Calcium 1.917 ( 2.48e+00 2.27e-01) 1.32e+00 ( 2.18e+00 1.45e-01)
368 oxide
369 00 11 R MNcn Log RI: refractive i 1.917 (-6.49e-01 1.18e-02) 7.03e-01 (-6.57e-01 5.82e-03)
370 ndex
371 00 14 R MNcn Log Al: Wt.% Aluminu 1.917 ( 6.73e-02 5.58e-01) 4.28e-01 ( 3.06e-01 3.70e-01)
372 m oxide
373 00 16 R MNcn Log K: Wt.% Potassi 1.917 (-2.29e+00 2.32e+00) 3.50e-01 (-1.47e+00 1.74e+00)
374 um oxide
375 00 12 R MNcn Log Na: Wt.% Sodium 1.917 ( 2.56e+00 1.27e-01) 2.93e-01 ( 2.59e+00 6.09e-02)
376 oxide
377 00 18 R MNcn Log Ba: Wt.% Barium 1.917 (-4.91e+00 1.67e+00) 2.88e-01 (-4.43e+00 1.93e+00)
378 oxide
379 00 15 R MNcn Log Si: Wt.% Silicon 1.917 ( 4.29e+00 2.25e-02) 9.96e-02 ( 4.29e+00 1.07e-02)
380 oxide
381 00 19 R MNcn Log Fe: Wt.% Iron ox 1.917 (-4.24e+00 1.83e+00) 3.37e-02 (-4.18e+00 1.66e+00)
382 ide
383
384 Correlation matrix (row & column indices are attribute numbers)
385 11 12 13 14 15 16 17 18 19
386 11 1.000 -0.618 0.000 0.252 -0.834 0.044 0.862 0.351 0.564
387 12 -0.618 1.000 0.000 -0.477 0.408 -0.279 -0.634 -0.369 -0.265
388 13 0.000 0.000 1.000 0.000 0.000 0.000 0.000 0.000 0.000
389 14 0.252 -0.477 0.000 1.000 -0.430 0.814 0.278 0.316 0.352
390 15 -0.834 0.408 0.000 -0.430 1.000 -0.249 -0.716 -0.479 -0.681
391 16 0.044 -0.279 0.000 0.814 -0.249 1.000 -0.002 0.196 0.294
392 17 0.862 -0.634 0.000 0.278 -0.716 -0.002 1.000 0.114 0.368
393 18 0.351 -0.369 0.000 0.316 -0.479 0.196 0.114 1.000 0.418
394 19 0.564 -0.265 0.000 0.352 -0.681 0.294 0.368 0.418 1.000
395