see5-installer 0.1.0
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +7 -0
- data/.gitignore +8 -0
- data/.rubocop.yml +11 -0
- data/CHANGELOG.md +5 -0
- data/Gemfile +10 -0
- data/README.md +29 -0
- data/Rakefile +12 -0
- data/ext/c5.0/Makefile +86 -0
- data/ext/c5.0/attwinnow.c +394 -0
- data/ext/c5.0/c50.c +330 -0
- data/ext/c5.0/classify.c +700 -0
- data/ext/c5.0/confmat.c +195 -0
- data/ext/c5.0/construct.c +853 -0
- data/ext/c5.0/contin.c +613 -0
- data/ext/c5.0/defns.i +788 -0
- data/ext/c5.0/discr.c +307 -0
- data/ext/c5.0/extern.i +170 -0
- data/ext/c5.0/formrules.c +720 -0
- data/ext/c5.0/formtree.c +1158 -0
- data/ext/c5.0/getdata.c +521 -0
- data/ext/c5.0/getnames.c +733 -0
- data/ext/c5.0/global.c +211 -0
- data/ext/c5.0/gpl.txt +674 -0
- data/ext/c5.0/implicitatt.c +1112 -0
- data/ext/c5.0/info.c +146 -0
- data/ext/c5.0/mcost.c +138 -0
- data/ext/c5.0/modelfiles.c +952 -0
- data/ext/c5.0/p-thresh.c +313 -0
- data/ext/c5.0/prune.c +1069 -0
- data/ext/c5.0/report.c +345 -0
- data/ext/c5.0/rules.c +579 -0
- data/ext/c5.0/ruletree.c +398 -0
- data/ext/c5.0/siftrules.c +1285 -0
- data/ext/c5.0/sort.c +156 -0
- data/ext/c5.0/subset.c +599 -0
- data/ext/c5.0/text.i +223 -0
- data/ext/c5.0/trees.c +740 -0
- data/ext/c5.0/update.c +129 -0
- data/ext/c5.0/utility.c +1146 -0
- data/ext/c5.0/xval +150 -0
- data/ext/c5.0/xval.c +402 -0
- data/ext/gritbot/Makefile +98 -0
- data/ext/gritbot/check.c +1110 -0
- data/ext/gritbot/cluster.c +342 -0
- data/ext/gritbot/common.c +1269 -0
- data/ext/gritbot/continatt.c +412 -0
- data/ext/gritbot/defns.i +623 -0
- data/ext/gritbot/discratt.c +459 -0
- data/ext/gritbot/extern.i +101 -0
- data/ext/gritbot/getdata.c +329 -0
- data/ext/gritbot/getnames.c +573 -0
- data/ext/gritbot/global.c +104 -0
- data/ext/gritbot/gpl.txt +674 -0
- data/ext/gritbot/gritbot.c +295 -0
- data/ext/gritbot/implicitatt.c +1108 -0
- data/ext/gritbot/inspect.c +794 -0
- data/ext/gritbot/modelfiles.c +687 -0
- data/ext/gritbot/outlier.c +415 -0
- data/ext/gritbot/sort.c +130 -0
- data/ext/gritbot/text.i +159 -0
- data/ext/gritbot/update.c +126 -0
- data/ext/gritbot/utility.c +1029 -0
- data/ext/see5-installer/extconf.rb +25 -0
- data/lib/see5/installer.rb +10 -0
- data/lib/see5/installer/version.rb +7 -0
- data/see5-installer.gemspec +30 -0
- metadata +115 -0
data/ext/c5.0/contin.c
ADDED
@@ -0,0 +1,613 @@
|
|
1
|
+
/*************************************************************************/
|
2
|
+
/* */
|
3
|
+
/* Copyright 2010 Rulequest Research Pty Ltd. */
|
4
|
+
/* */
|
5
|
+
/* This file is part of C5.0 GPL Edition, a single-threaded version */
|
6
|
+
/* of C5.0 release 2.07. */
|
7
|
+
/* */
|
8
|
+
/* C5.0 GPL Edition is free software: you can redistribute it and/or */
|
9
|
+
/* modify it under the terms of the GNU General Public License as */
|
10
|
+
/* published by the Free Software Foundation, either version 3 of the */
|
11
|
+
/* License, or (at your option) any later version. */
|
12
|
+
/* */
|
13
|
+
/* C5.0 GPL Edition is distributed in the hope that it will be useful, */
|
14
|
+
/* but WITHOUT ANY WARRANTY; without even the implied warranty of */
|
15
|
+
/* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU */
|
16
|
+
/* General Public License for more details. */
|
17
|
+
/* */
|
18
|
+
/* You should have received a copy of the GNU General Public License */
|
19
|
+
/* (gpl.txt) along with C5.0 GPL Edition. If not, see */
|
20
|
+
/* */
|
21
|
+
/* <http://www.gnu.org/licenses/>. */
|
22
|
+
/* */
|
23
|
+
/*************************************************************************/
|
24
|
+
|
25
|
+
|
26
|
+
|
27
|
+
/*************************************************************************/
|
28
|
+
/* */
|
29
|
+
/* Evaluation of a test on a continuous valued attribute */
|
30
|
+
/* ----------------------------------------------------- */
|
31
|
+
/* */
|
32
|
+
/*************************************************************************/
|
33
|
+
|
34
|
+
#include "defns.i"
|
35
|
+
#include "extern.i"
|
36
|
+
|
37
|
+
#define PartInfo(n) (-(n)*Log((n)/GEnv.Cases))
|
38
|
+
|
39
|
+
|
40
|
+
/*************************************************************************/
|
41
|
+
/* */
|
42
|
+
/* Continuous attributes are treated as if they have possible */
|
43
|
+
/* values 0 (unknown), 1 (not applicable), 2 (less than cut) and */
|
44
|
+
/* 3 (greater than cut). */
|
45
|
+
/* This routine finds the best cut for cases Fp through Lp and */
|
46
|
+
/* sets Info[], Gain[] and Bar[] */
|
47
|
+
/* */
|
48
|
+
/*************************************************************************/
|
49
|
+
|
50
|
+
|
51
|
+
void EvalContinuousAtt(Attribute Att, CaseNo Fp, CaseNo Lp)
|
52
|
+
/* ----------------- */
|
53
|
+
{
|
54
|
+
CaseNo i, j, BestI, Tries=0;
|
55
|
+
double LowInfo, LHInfo, LeastInfo=1E38,
|
56
|
+
w, BestGain, BestInfo, ThreshCost=1;
|
57
|
+
ClassNo c;
|
58
|
+
ContValue Interval;
|
59
|
+
|
60
|
+
Verbosity(3, fprintf(Of, "\tAtt %s\n", AttName[Att]))
|
61
|
+
|
62
|
+
Gain[Att] = None;
|
63
|
+
PrepareForContin(Att, Fp, Lp);
|
64
|
+
|
65
|
+
/* Special case when very few known values */
|
66
|
+
|
67
|
+
if ( GEnv.ApplicCases < 2 * MINITEMS )
|
68
|
+
{
|
69
|
+
Verbosity(2,
|
70
|
+
fprintf(Of, "\tAtt %s\tinsufficient cases with known values\n",
|
71
|
+
AttName[Att]))
|
72
|
+
return;
|
73
|
+
}
|
74
|
+
|
75
|
+
/* Try possible cuts between cases i and i+1, and determine the
|
76
|
+
information and gain of the split in each case */
|
77
|
+
|
78
|
+
/* We have to be wary of splitting a small number of cases off one end,
|
79
|
+
as this has little predictive power. The minimum split GEnv.MinSplit is
|
80
|
+
the maximum of MINITEMS or (the minimum of 25 and 10% of the cases
|
81
|
+
per class) */
|
82
|
+
|
83
|
+
GEnv.MinSplit = 0.10 * GEnv.KnownCases / MaxClass;
|
84
|
+
if ( GEnv.MinSplit > 25 ) GEnv.MinSplit = 25;
|
85
|
+
if ( GEnv.MinSplit < MINITEMS ) GEnv.MinSplit = MINITEMS;
|
86
|
+
|
87
|
+
/* Find first possible cut point and initialise scan parameters */
|
88
|
+
|
89
|
+
i = PrepareForScan(Lp);
|
90
|
+
|
91
|
+
/* Repeatedly check next possible cut */
|
92
|
+
|
93
|
+
for ( ; i <= GEnv.Ep ; i++ )
|
94
|
+
{
|
95
|
+
c = GEnv.SRec[i].C;
|
96
|
+
w = GEnv.SRec[i].W;
|
97
|
+
assert(c >= 1 && c <= MaxClass);
|
98
|
+
|
99
|
+
GEnv.LowCases += w;
|
100
|
+
GEnv.Freq[2][c] += w;
|
101
|
+
GEnv.Freq[3][c] -= w;
|
102
|
+
|
103
|
+
GEnv.HighVal = GEnv.SRec[i+1].V;
|
104
|
+
if ( GEnv.HighVal > GEnv.LowVal )
|
105
|
+
{
|
106
|
+
Tries++;
|
107
|
+
|
108
|
+
GEnv.LowClass = GEnv.HighClass;
|
109
|
+
GEnv.HighClass = GEnv.SRec[i+1].C;
|
110
|
+
for ( j = i+2 ;
|
111
|
+
GEnv.HighClass && j <= GEnv.Ep && GEnv.SRec[j].V == GEnv.HighVal ;
|
112
|
+
j++ )
|
113
|
+
{
|
114
|
+
if ( GEnv.SRec[j].C != GEnv.HighClass ) GEnv.HighClass = 0;
|
115
|
+
}
|
116
|
+
|
117
|
+
if ( ! GEnv.LowClass || GEnv.LowClass != GEnv.HighClass || j > GEnv.Ep )
|
118
|
+
{
|
119
|
+
LowInfo = TotalInfo(GEnv.Freq[2], 1, MaxClass);
|
120
|
+
|
121
|
+
/* If cannot improve on best so far, count remaining
|
122
|
+
possible cuts and break */
|
123
|
+
|
124
|
+
if ( LowInfo >= LeastInfo )
|
125
|
+
{
|
126
|
+
for ( i++ ; i <= GEnv.Ep ; i++ )
|
127
|
+
{
|
128
|
+
if ( GEnv.SRec[i+1].V > GEnv.SRec[i].V )
|
129
|
+
{
|
130
|
+
Tries++;
|
131
|
+
}
|
132
|
+
}
|
133
|
+
break;
|
134
|
+
}
|
135
|
+
|
136
|
+
LHInfo = LowInfo + TotalInfo(GEnv.Freq[3], 1, MaxClass);
|
137
|
+
if ( LHInfo < LeastInfo )
|
138
|
+
{
|
139
|
+
LeastInfo = LHInfo;
|
140
|
+
BestI = i;
|
141
|
+
|
142
|
+
BestInfo = (GEnv.FixedSplitInfo
|
143
|
+
+ PartInfo(GEnv.LowCases)
|
144
|
+
+ PartInfo(GEnv.ApplicCases - GEnv.LowCases))
|
145
|
+
/ GEnv.Cases;
|
146
|
+
}
|
147
|
+
|
148
|
+
Verbosity(3,
|
149
|
+
{
|
150
|
+
fprintf(Of, "\t\tCut at %.3f (gain %.3f):",
|
151
|
+
(GEnv.LowVal + GEnv.HighVal) / 2,
|
152
|
+
(1 - GEnv.UnknownRate) *
|
153
|
+
(GEnv.BaseInfo - (GEnv.NAInfo + LHInfo) / GEnv.KnownCases));
|
154
|
+
PrintDistribution(Att, 2, 3, GEnv.Freq, GEnv.ValFreq, true);
|
155
|
+
})
|
156
|
+
}
|
157
|
+
|
158
|
+
GEnv.LowVal = GEnv.HighVal;
|
159
|
+
}
|
160
|
+
}
|
161
|
+
|
162
|
+
BestGain = (1 - GEnv.UnknownRate) *
|
163
|
+
(GEnv.BaseInfo - (GEnv.NAInfo + LeastInfo) / GEnv.KnownCases);
|
164
|
+
|
165
|
+
/* The threshold cost is the lesser of the cost of indicating the
|
166
|
+
cases to split between or the interval containing the split */
|
167
|
+
|
168
|
+
if ( BestGain > 0 )
|
169
|
+
{
|
170
|
+
Interval = (GEnv.SRec[Lp].V - GEnv.SRec[GEnv.Xp].V) /
|
171
|
+
(GEnv.SRec[BestI+1].V - GEnv.SRec[BestI].V);
|
172
|
+
ThreshCost = ( Interval < Tries ? Log(Interval) : Log(Tries) )
|
173
|
+
/ GEnv.Cases;
|
174
|
+
}
|
175
|
+
|
176
|
+
BestGain -= ThreshCost;
|
177
|
+
|
178
|
+
/* If a test on the attribute is able to make a gain,
|
179
|
+
set the best break point, gain and information */
|
180
|
+
|
181
|
+
if ( BestGain <= 0 )
|
182
|
+
{
|
183
|
+
Verbosity(2, fprintf(Of, "\tAtt %s\tno gain\n", AttName[Att]))
|
184
|
+
}
|
185
|
+
else
|
186
|
+
{
|
187
|
+
Gain[Att] = BestGain;
|
188
|
+
Info[Att] = BestInfo;
|
189
|
+
|
190
|
+
GEnv.LowVal = GEnv.SRec[BestI].V;
|
191
|
+
GEnv.HighVal = GEnv.SRec[BestI+1].V;
|
192
|
+
|
193
|
+
/* Set threshold, making sure that rounding problems do not
|
194
|
+
cause it to reach upper value */
|
195
|
+
|
196
|
+
if ( (Bar[Att] = (ContValue) (0.5 * (GEnv.LowVal + GEnv.HighVal)))
|
197
|
+
>= GEnv.HighVal )
|
198
|
+
{
|
199
|
+
Bar[Att] = GEnv.LowVal;
|
200
|
+
}
|
201
|
+
|
202
|
+
Verbosity(2,
|
203
|
+
fprintf(Of, "\tAtt %s\tcut=%.3f, inf %.3f, gain %.3f\n",
|
204
|
+
AttName[Att], Bar[Att], Info[Att], Gain[Att]))
|
205
|
+
}
|
206
|
+
}
|
207
|
+
|
208
|
+
|
209
|
+
|
210
|
+
/*************************************************************************/
|
211
|
+
/* */
|
212
|
+
/* Estimate max gain ratio available from any cut, using sample */
|
213
|
+
/* of SampleFrac of all cases */
|
214
|
+
/* */
|
215
|
+
/*************************************************************************/
|
216
|
+
|
217
|
+
|
218
|
+
void EstimateMaxGR(Attribute Att, CaseNo Fp, CaseNo Lp)
|
219
|
+
/* ------------- */
|
220
|
+
{
|
221
|
+
CaseNo i, j;
|
222
|
+
double LHInfo, w, SplitInfo, ThisGain, GR;
|
223
|
+
ClassNo c;
|
224
|
+
|
225
|
+
EstMaxGR[Att] = 0;
|
226
|
+
|
227
|
+
if ( Skip(Att) || Att == ClassAtt ) return;
|
228
|
+
|
229
|
+
PrepareForContin(Att, Fp, Lp);
|
230
|
+
|
231
|
+
/* Special case when very few known values */
|
232
|
+
|
233
|
+
if ( GEnv.ApplicCases < 2 * MINITEMS * SampleFrac )
|
234
|
+
{
|
235
|
+
return;
|
236
|
+
}
|
237
|
+
|
238
|
+
/* Try possible cuts between cases i and i+1. Use conservative
|
239
|
+
value of GEnv.MinSplit to allow for sampling */
|
240
|
+
|
241
|
+
GEnv.MinSplit = 0.10 * GEnv.KnownCases / MaxClass;
|
242
|
+
if ( GEnv.MinSplit > 25 ) GEnv.MinSplit = 25;
|
243
|
+
if ( GEnv.MinSplit < MINITEMS ) GEnv.MinSplit = MINITEMS;
|
244
|
+
|
245
|
+
GEnv.MinSplit *= SampleFrac * 0.33;
|
246
|
+
|
247
|
+
i = PrepareForScan(Lp);
|
248
|
+
|
249
|
+
/* Repeatedly check next possible cut */
|
250
|
+
|
251
|
+
for ( ; i <= GEnv.Ep ; i++ )
|
252
|
+
{
|
253
|
+
c = GEnv.SRec[i].C;
|
254
|
+
w = GEnv.SRec[i].W;
|
255
|
+
assert(c >= 1 && c <= MaxClass);
|
256
|
+
|
257
|
+
GEnv.LowCases += w;
|
258
|
+
GEnv.Freq[2][c] += w;
|
259
|
+
GEnv.Freq[3][c] -= w;
|
260
|
+
|
261
|
+
GEnv.HighVal = GEnv.SRec[i+1].V;
|
262
|
+
if ( GEnv.HighVal > GEnv.LowVal )
|
263
|
+
{
|
264
|
+
GEnv.LowClass = GEnv.HighClass;
|
265
|
+
GEnv.HighClass = GEnv.SRec[i+1].C;
|
266
|
+
for ( j = i+2 ;
|
267
|
+
GEnv.HighClass && j <= GEnv.Ep && GEnv.SRec[j].V == GEnv.HighVal ;
|
268
|
+
j++ )
|
269
|
+
{
|
270
|
+
if ( GEnv.SRec[j].C != GEnv.HighClass ) GEnv.HighClass = 0;
|
271
|
+
}
|
272
|
+
|
273
|
+
if ( ! GEnv.LowClass || GEnv.LowClass != GEnv.HighClass || j > GEnv.Ep )
|
274
|
+
{
|
275
|
+
LHInfo = TotalInfo(GEnv.Freq[2], 1, MaxClass)
|
276
|
+
+ TotalInfo(GEnv.Freq[3], 1, MaxClass);
|
277
|
+
|
278
|
+
SplitInfo = (GEnv.FixedSplitInfo
|
279
|
+
+ PartInfo(GEnv.LowCases)
|
280
|
+
+ PartInfo(GEnv.ApplicCases - GEnv.LowCases)) / GEnv.Cases;
|
281
|
+
|
282
|
+
ThisGain = (1 - GEnv.UnknownRate) *
|
283
|
+
(GEnv.BaseInfo - (GEnv.NAInfo + LHInfo) / GEnv.KnownCases);
|
284
|
+
if ( ThisGain > Gain[Att] ) Gain[Att] = ThisGain;
|
285
|
+
|
286
|
+
/* Adjust GR to make it more conservative upper bound */
|
287
|
+
|
288
|
+
GR = (ThisGain + 1E-5) / SplitInfo;
|
289
|
+
if ( GR > EstMaxGR[Att] )
|
290
|
+
{
|
291
|
+
EstMaxGR[Att] = GR;
|
292
|
+
}
|
293
|
+
|
294
|
+
Verbosity(3,
|
295
|
+
{
|
296
|
+
fprintf(Of, "\t\tCut at %.3f (gain %.3f):",
|
297
|
+
(GEnv.LowVal + GEnv.HighVal) / 2, ThisGain);
|
298
|
+
PrintDistribution(Att, 2, 3, GEnv.Freq, GEnv.ValFreq, true);
|
299
|
+
})
|
300
|
+
}
|
301
|
+
|
302
|
+
GEnv.LowVal = GEnv.HighVal;
|
303
|
+
}
|
304
|
+
}
|
305
|
+
|
306
|
+
Verbosity(2,
|
307
|
+
fprintf(Of, "\tAtt %s: max GR estimate %.3f\n",
|
308
|
+
AttName[Att], EstMaxGR[Att]))
|
309
|
+
}
|
310
|
+
|
311
|
+
|
312
|
+
|
313
|
+
/*************************************************************************/
|
314
|
+
/* */
|
315
|
+
/* Routine to set some preparatory values used by both */
|
316
|
+
/* EvalContinuousAtt and EstimateMaxGR */
|
317
|
+
/* */
|
318
|
+
/*************************************************************************/
|
319
|
+
|
320
|
+
|
321
|
+
void PrepareForContin(Attribute Att, CaseNo Fp, CaseNo Lp)
|
322
|
+
/* ---------------- */
|
323
|
+
{
|
324
|
+
CaseNo i;
|
325
|
+
ClassNo c;
|
326
|
+
DiscrValue v;
|
327
|
+
|
328
|
+
/* Reset frequency tables */
|
329
|
+
|
330
|
+
ForEach(v, 0, 3)
|
331
|
+
{
|
332
|
+
ForEach(c, 1, MaxClass)
|
333
|
+
{
|
334
|
+
GEnv.Freq[v][c] = 0;
|
335
|
+
}
|
336
|
+
GEnv.ValFreq[v] = 0;
|
337
|
+
}
|
338
|
+
|
339
|
+
/* Omit and count unknown and N/A values */
|
340
|
+
|
341
|
+
GEnv.Cases = 0;
|
342
|
+
|
343
|
+
if ( SomeMiss[Att] || SomeNA[Att] )
|
344
|
+
{
|
345
|
+
GEnv.Xp = Lp+1;
|
346
|
+
|
347
|
+
ForEach(i, Fp, Lp)
|
348
|
+
{
|
349
|
+
assert(Class(Case[i]) >= 1 && Class(Case[i]) <= MaxClass);
|
350
|
+
|
351
|
+
GEnv.Cases += Weight(Case[i]);
|
352
|
+
|
353
|
+
if ( Unknown(Case[i], Att) )
|
354
|
+
{
|
355
|
+
GEnv.Freq[ 0 ][ Class(Case[i]) ] += Weight(Case[i]);
|
356
|
+
}
|
357
|
+
else
|
358
|
+
if ( NotApplic(Case[i], Att) )
|
359
|
+
{
|
360
|
+
GEnv.Freq[ 1 ][ Class(Case[i]) ] += Weight(Case[i]);
|
361
|
+
}
|
362
|
+
else
|
363
|
+
{
|
364
|
+
GEnv.Freq[ 3 ][ Class(Case[i]) ] += Weight(Case[i]);
|
365
|
+
GEnv.Xp--;
|
366
|
+
GEnv.SRec[GEnv.Xp].V = CVal(Case[i], Att);
|
367
|
+
GEnv.SRec[GEnv.Xp].W = Weight(Case[i]);
|
368
|
+
GEnv.SRec[GEnv.Xp].C = Class(Case[i]);
|
369
|
+
}
|
370
|
+
}
|
371
|
+
|
372
|
+
ForEach(c, 1, MaxClass)
|
373
|
+
{
|
374
|
+
GEnv.ValFreq[0] += GEnv.Freq[0][c];
|
375
|
+
GEnv.ValFreq[1] += GEnv.Freq[1][c];
|
376
|
+
}
|
377
|
+
|
378
|
+
GEnv.NAInfo = TotalInfo(GEnv.Freq[1], 1, MaxClass);
|
379
|
+
GEnv.FixedSplitInfo = PartInfo(GEnv.ValFreq[0]) + PartInfo(GEnv.ValFreq[1]);
|
380
|
+
|
381
|
+
Verbosity(3, PrintDistribution(Att, 0, 1, GEnv.Freq, GEnv.ValFreq, true))
|
382
|
+
}
|
383
|
+
else
|
384
|
+
{
|
385
|
+
GEnv.Xp = Fp;
|
386
|
+
|
387
|
+
ForEach(i, Fp, Lp)
|
388
|
+
{
|
389
|
+
GEnv.SRec[i].V = CVal(Case[i], Att);
|
390
|
+
GEnv.SRec[i].W = Weight(Case[i]);
|
391
|
+
GEnv.SRec[i].C = Class(Case[i]);
|
392
|
+
|
393
|
+
GEnv.Freq[3][Class(Case[i])] += Weight(Case[i]);
|
394
|
+
}
|
395
|
+
|
396
|
+
ForEach(c, 1, MaxClass)
|
397
|
+
{
|
398
|
+
GEnv.Cases += GEnv.Freq[3][c];
|
399
|
+
}
|
400
|
+
|
401
|
+
GEnv.NAInfo = GEnv.FixedSplitInfo = 0;
|
402
|
+
}
|
403
|
+
|
404
|
+
GEnv.KnownCases = GEnv.Cases - GEnv.ValFreq[0];
|
405
|
+
GEnv.ApplicCases = GEnv.KnownCases - GEnv.ValFreq[1];
|
406
|
+
|
407
|
+
GEnv.UnknownRate = 1.0 - GEnv.KnownCases / GEnv.Cases;
|
408
|
+
|
409
|
+
Cachesort(GEnv.Xp, Lp, GEnv.SRec);
|
410
|
+
|
411
|
+
/* If unknowns or using sampling, must recompute base information */
|
412
|
+
|
413
|
+
if ( GEnv.ValFreq[0] > 0 || SampleFrac < 1 )
|
414
|
+
{
|
415
|
+
/* Determine base information using GEnv.Freq[0] as temp buffer */
|
416
|
+
|
417
|
+
ForEach(c, 1, MaxClass)
|
418
|
+
{
|
419
|
+
GEnv.Freq[0][c] = GEnv.Freq[1][c] + GEnv.Freq[3][c];
|
420
|
+
}
|
421
|
+
|
422
|
+
GEnv.BaseInfo = TotalInfo(GEnv.Freq[0], 1, MaxClass) / GEnv.KnownCases;
|
423
|
+
}
|
424
|
+
else
|
425
|
+
{
|
426
|
+
GEnv.BaseInfo = GlobalBaseInfo;
|
427
|
+
}
|
428
|
+
}
|
429
|
+
|
430
|
+
|
431
|
+
|
432
|
+
/*************************************************************************/
|
433
|
+
/* */
|
434
|
+
/* Set low and high bounds for scan and initial class */
|
435
|
+
/* (used by EvalContinuousAtt and EstimateMaxGR) */
|
436
|
+
/* */
|
437
|
+
/*************************************************************************/
|
438
|
+
|
439
|
+
|
440
|
+
CaseNo PrepareForScan(CaseNo Lp)
|
441
|
+
/* -------------- */
|
442
|
+
{
|
443
|
+
CaseNo i, j;
|
444
|
+
ClassNo c;
|
445
|
+
double w;
|
446
|
+
|
447
|
+
/* Find last possible split */
|
448
|
+
|
449
|
+
GEnv.HighCases = GEnv.LowCases = 0;
|
450
|
+
|
451
|
+
for ( GEnv.Ep = Lp ; GEnv.Ep >= GEnv.Xp && GEnv.HighCases < GEnv.MinSplit ; GEnv.Ep-- )
|
452
|
+
{
|
453
|
+
GEnv.HighCases += GEnv.SRec[GEnv.Ep].W;
|
454
|
+
}
|
455
|
+
|
456
|
+
/* Skip cases before first possible cut */
|
457
|
+
|
458
|
+
for ( i = GEnv.Xp ;
|
459
|
+
i <= GEnv.Ep &&
|
460
|
+
( GEnv.LowCases + GEnv.SRec[i].W < GEnv.MinSplit - 1E-5 ||
|
461
|
+
GEnv.SRec[i].V == GEnv.SRec[i+1].V ) ;
|
462
|
+
i++ )
|
463
|
+
{
|
464
|
+
c = GEnv.SRec[i].C;
|
465
|
+
w = GEnv.SRec[i].W;
|
466
|
+
assert(c >= 1 && c <= MaxClass);
|
467
|
+
|
468
|
+
GEnv.LowCases += w;
|
469
|
+
GEnv.Freq[2][c] += w;
|
470
|
+
GEnv.Freq[3][c] -= w;
|
471
|
+
}
|
472
|
+
|
473
|
+
/* Find the class key for the first interval */
|
474
|
+
|
475
|
+
GEnv.HighClass = GEnv.SRec[i].C;
|
476
|
+
for ( j = i-1; GEnv.HighClass && j >= GEnv.Xp ; j-- )
|
477
|
+
{
|
478
|
+
if ( GEnv.SRec[j].C != GEnv.HighClass ) GEnv.HighClass = 0;
|
479
|
+
}
|
480
|
+
assert(GEnv.HighClass <= MaxClass);
|
481
|
+
assert(j+1 >= GEnv.Xp);
|
482
|
+
|
483
|
+
GEnv.LowVal = GEnv.SRec[i].V;
|
484
|
+
|
485
|
+
return i;
|
486
|
+
}
|
487
|
+
|
488
|
+
|
489
|
+
|
490
|
+
/*************************************************************************/
|
491
|
+
/* */
|
492
|
+
/* Change a leaf into a test on a continuous attribute */
|
493
|
+
/* */
|
494
|
+
/*************************************************************************/
|
495
|
+
|
496
|
+
|
497
|
+
void ContinTest(Tree Node, Attribute Att)
|
498
|
+
/* ---------- */
|
499
|
+
{
|
500
|
+
Sprout(Node, 3);
|
501
|
+
|
502
|
+
Node->NodeType = BrThresh;
|
503
|
+
Node->Tested = Att;
|
504
|
+
Node->Cut =
|
505
|
+
Node->Lower =
|
506
|
+
Node->Upper = Bar[Att];
|
507
|
+
}
|
508
|
+
|
509
|
+
|
510
|
+
|
511
|
+
/*************************************************************************/
|
512
|
+
/* */
|
513
|
+
/* Adjust thresholds of all continuous attributes so that cuts */
|
514
|
+
/* are values that appear in the data */
|
515
|
+
/* */
|
516
|
+
/*************************************************************************/
|
517
|
+
|
518
|
+
|
519
|
+
void AdjustAllThresholds(Tree T)
|
520
|
+
/* ------------------- */
|
521
|
+
{
|
522
|
+
Attribute Att;
|
523
|
+
CaseNo Ep;
|
524
|
+
|
525
|
+
ForEach(Att, 1, MaxAtt)
|
526
|
+
{
|
527
|
+
if ( Continuous(Att) )
|
528
|
+
{
|
529
|
+
Ep = -1;
|
530
|
+
AdjustThresholds(T, Att, &Ep);
|
531
|
+
}
|
532
|
+
}
|
533
|
+
}
|
534
|
+
|
535
|
+
|
536
|
+
|
537
|
+
void AdjustThresholds(Tree T, Attribute Att, CaseNo *Ep)
|
538
|
+
/* ---------------- */
|
539
|
+
{
|
540
|
+
DiscrValue v;
|
541
|
+
CaseNo i;
|
542
|
+
|
543
|
+
if ( T->NodeType == BrThresh && T->Tested == Att )
|
544
|
+
{
|
545
|
+
if ( *Ep == -1 )
|
546
|
+
{
|
547
|
+
ForEach(i, 0, MaxCase)
|
548
|
+
{
|
549
|
+
if ( ! Unknown(Case[i], Att) && ! NotApplic(Case[i], Att) )
|
550
|
+
{
|
551
|
+
(&GEnv)->SRec[++(*Ep)].V = CVal(Case[i], Att);
|
552
|
+
}
|
553
|
+
}
|
554
|
+
Cachesort(0, *Ep, (&GEnv)->SRec);
|
555
|
+
|
556
|
+
if ( PossibleCuts && Trial == 0 )
|
557
|
+
{
|
558
|
+
int Cuts=0;
|
559
|
+
|
560
|
+
ForEach(i, 1, *Ep)
|
561
|
+
{
|
562
|
+
if ( (&GEnv)->SRec[i].V != (&GEnv)->SRec[i-1].V ) Cuts++;
|
563
|
+
}
|
564
|
+
PossibleCuts[Att] = Cuts;
|
565
|
+
}
|
566
|
+
}
|
567
|
+
|
568
|
+
T->Cut = T->Lower = T->Upper = GreatestValueBelow(T->Cut, Ep);
|
569
|
+
}
|
570
|
+
|
571
|
+
if ( T->NodeType )
|
572
|
+
{
|
573
|
+
ForEach(v, 1, T->Forks)
|
574
|
+
{
|
575
|
+
AdjustThresholds(T->Branch[v], Att, Ep);
|
576
|
+
}
|
577
|
+
}
|
578
|
+
}
|
579
|
+
|
580
|
+
|
581
|
+
|
582
|
+
/*************************************************************************/
|
583
|
+
/* */
|
584
|
+
/* Return the greatest value of attribute Att below threshold Th */
|
585
|
+
/* (Assumes values of Att have been sorted.) */
|
586
|
+
/* */
|
587
|
+
/*************************************************************************/
|
588
|
+
|
589
|
+
|
590
|
+
ContValue GreatestValueBelow(ContValue Th, CaseNo *Ep)
|
591
|
+
/* ------------------ */
|
592
|
+
{
|
593
|
+
CaseNo Low, Mid, High;
|
594
|
+
|
595
|
+
Low = 0;
|
596
|
+
High = *Ep;
|
597
|
+
|
598
|
+
while ( Low < High )
|
599
|
+
{
|
600
|
+
Mid = (Low + High + 1) / 2;
|
601
|
+
|
602
|
+
if ( (&GEnv)->SRec[Mid].V > Th )
|
603
|
+
{
|
604
|
+
High = Mid - 1;
|
605
|
+
}
|
606
|
+
else
|
607
|
+
{
|
608
|
+
Low = Mid;
|
609
|
+
}
|
610
|
+
}
|
611
|
+
|
612
|
+
return (&GEnv)->SRec[Low].V;
|
613
|
+
}
|