1 | /*************************************************************************/ |
---|
2 | /* */ |
---|
3 | /* Classify items interactively using a set of rules */ |
---|
4 | /* ------------------------------------------------- */ |
---|
5 | /* */ |
---|
6 | /*************************************************************************/ |
---|
7 | |
---|
8 | |
---|
9 | #include "defns.i" |
---|
10 | #include "types.i" |
---|
11 | |
---|
12 | |
---|
13 | /* External data */ |
---|
14 | |
---|
15 | short MaxAtt, MaxClass, MaxDiscrVal; |
---|
16 | |
---|
17 | ItemNo MaxItem; |
---|
18 | |
---|
19 | Description *Item; |
---|
20 | |
---|
21 | DiscrValue *MaxAttVal; |
---|
22 | |
---|
23 | String *ClassName, |
---|
24 | *AttName, |
---|
25 | **AttValName, |
---|
26 | FileName = "DF"; |
---|
27 | |
---|
28 | char *SpecialStatus; |
---|
29 | |
---|
30 | short VERBOSITY = 0, |
---|
31 | TRACE = 0; |
---|
32 | |
---|
33 | |
---|
34 | Boolean FirstTime = true; |
---|
35 | |
---|
36 | PR *Rule; |
---|
37 | |
---|
38 | RuleNo NRules = 0, |
---|
39 | *RuleIndex; |
---|
40 | |
---|
41 | short RuleSpace = 0; |
---|
42 | |
---|
43 | ClassNo DefaultClass; |
---|
44 | |
---|
45 | |
---|
46 | /* The interview module uses a more complex description of an |
---|
47 | case called a "Range Description" with lower and upper bounds. |
---|
48 | For further information, see consult.c */ |
---|
49 | |
---|
50 | typedef struct ValRange *RangeDescRec; |
---|
51 | |
---|
52 | struct ValRange |
---|
53 | { |
---|
54 | Boolean Known, Asked; |
---|
55 | float LowerBound, UpperBound, *Probability; |
---|
56 | }; |
---|
57 | |
---|
58 | RangeDescRec RangeDesc; |
---|
59 | |
---|
60 | float Confidence; |
---|
61 | |
---|
62 | #define MINCF 0.50 /* minimum cf for useable rule */ |
---|
63 | |
---|
64 | |
---|
65 | /*************************************************************************/ |
---|
66 | /* */ |
---|
67 | /* Find the best rule for the current case. */ |
---|
68 | /* Note: leave probability in Confidence */ |
---|
69 | /* */ |
---|
70 | /*************************************************************************/ |
---|
71 | |
---|
72 | |
---|
73 | RuleNo BestRule() |
---|
74 | /* -------- */ |
---|
75 | { |
---|
76 | RuleNo r; |
---|
77 | float cf, RuleStrength(); |
---|
78 | |
---|
79 | Confidence = 0.0; |
---|
80 | |
---|
81 | ForEach(r, 1, NRules) |
---|
82 | { |
---|
83 | cf = RuleStrength(Rule[r]); |
---|
84 | |
---|
85 | if ( cf > 0.3 ) |
---|
86 | { |
---|
87 | Confidence = cf; |
---|
88 | return r; |
---|
89 | } |
---|
90 | } |
---|
91 | |
---|
92 | return 0; |
---|
93 | } |
---|
94 | |
---|
95 | |
---|
96 | |
---|
97 | /*************************************************************************/ |
---|
98 | /* */ |
---|
99 | /* Given a rule, determine the strength with which we can conclude */ |
---|
100 | /* that the current case belongs to the class specified by the RHS */ |
---|
101 | /* of the rule */ |
---|
102 | /* */ |
---|
103 | /*************************************************************************/ |
---|
104 | |
---|
105 | |
---|
106 | float RuleStrength(Rule) |
---|
107 | /* ------------ */ |
---|
108 | PR Rule; |
---|
109 | { |
---|
110 | short d; |
---|
111 | float RuleProb=1.0, ProbSatisfied(); |
---|
112 | |
---|
113 | ForEach(d, 1, Rule.Size) |
---|
114 | { |
---|
115 | RuleProb *= ProbSatisfied(Rule.Lhs[d]); |
---|
116 | if ( RuleProb < MINCF ) |
---|
117 | { |
---|
118 | return 0.0; |
---|
119 | } |
---|
120 | } |
---|
121 | |
---|
122 | return ( (1 - Rule.Error) * RuleProb ); |
---|
123 | } |
---|
124 | |
---|
125 | |
---|
126 | |
---|
127 | /*************************************************************************/ |
---|
128 | /* */ |
---|
129 | /* Determine the probability of the current case description */ |
---|
130 | /* satisfying the given condition */ |
---|
131 | /* */ |
---|
132 | /*************************************************************************/ |
---|
133 | |
---|
134 | |
---|
135 | float ProbSatisfied(c) |
---|
136 | /* ------------- */ |
---|
137 | Condition c; |
---|
138 | { |
---|
139 | Attribute a; |
---|
140 | char v; |
---|
141 | float AddProb=0.0; |
---|
142 | Test t; |
---|
143 | DiscrValue i; |
---|
144 | |
---|
145 | t = c->CondTest; |
---|
146 | a = t->Tested; |
---|
147 | v = c->TestValue; |
---|
148 | |
---|
149 | CheckValue(a, Nil); |
---|
150 | |
---|
151 | if ( ! RangeDesc[a].Known ) |
---|
152 | { |
---|
153 | return 0.0; |
---|
154 | } |
---|
155 | |
---|
156 | switch ( t->NodeType ) |
---|
157 | { |
---|
158 | case BrDiscr: /* test of discrete attribute */ |
---|
159 | |
---|
160 | return RangeDesc[a].Probability[v]; |
---|
161 | |
---|
162 | case ThreshContin: /* test of continuous attribute */ |
---|
163 | |
---|
164 | if ( RangeDesc[a].UpperBound <= t->Cut ) |
---|
165 | { |
---|
166 | return ( v == 1 ? 1.0 : 0.0 ); |
---|
167 | } |
---|
168 | else |
---|
169 | if ( RangeDesc[a].LowerBound > t->Cut ) |
---|
170 | { |
---|
171 | return ( v == 2 ? 1.0 : 0.0 ); |
---|
172 | } |
---|
173 | else |
---|
174 | if ( v == 1 ) |
---|
175 | { |
---|
176 | return (t->Cut - RangeDesc[a].LowerBound) / |
---|
177 | (RangeDesc[a].UpperBound - RangeDesc[a].LowerBound); |
---|
178 | } |
---|
179 | else |
---|
180 | { |
---|
181 | return (RangeDesc[a].UpperBound - t->Cut) / |
---|
182 | (RangeDesc[a].UpperBound - RangeDesc[a].LowerBound); |
---|
183 | } |
---|
184 | |
---|
185 | case BrSubset: /* subset test on discrete attribute */ |
---|
186 | |
---|
187 | ForEach(i, 1, MaxAttVal[a]) |
---|
188 | { |
---|
189 | if ( In(i, t->Subset[v]) ) |
---|
190 | { |
---|
191 | AddProb += RangeDesc[a].Probability[i]; |
---|
192 | } |
---|
193 | } |
---|
194 | return AddProb; |
---|
195 | |
---|
196 | } |
---|
197 | return 0.0; |
---|
198 | } |
---|
199 | |
---|
200 | |
---|
201 | |
---|
202 | /*************************************************************************/ |
---|
203 | /* */ |
---|
204 | /* Process a single case */ |
---|
205 | /* */ |
---|
206 | /*************************************************************************/ |
---|
207 | |
---|
208 | |
---|
209 | InterpretRuleset() |
---|
210 | /* ---------------- */ |
---|
211 | { |
---|
212 | char Reply; |
---|
213 | Attribute a; |
---|
214 | RuleNo r; |
---|
215 | |
---|
216 | /* Initialise */ |
---|
217 | |
---|
218 | ForEach(a, 0, MaxAtt) |
---|
219 | { |
---|
220 | RangeDesc[a].Asked = false; |
---|
221 | } |
---|
222 | |
---|
223 | if ( FirstTime ) |
---|
224 | { |
---|
225 | FirstTime = false; |
---|
226 | printf("\n"); |
---|
227 | } |
---|
228 | else |
---|
229 | { |
---|
230 | printf("\n-------------------------------------------\n\n"); |
---|
231 | } |
---|
232 | |
---|
233 | /* Find the first rule that fires on the item */ |
---|
234 | |
---|
235 | if ( r = BestRule() ) |
---|
236 | { |
---|
237 | printf("\nDecision:\n"); |
---|
238 | printf("\t%s", ClassName[Rule[r].Rhs]); |
---|
239 | |
---|
240 | if ( Confidence < 1.0 ) |
---|
241 | { |
---|
242 | printf(" CF = %.2f", Confidence); |
---|
243 | } |
---|
244 | |
---|
245 | printf("\n"); |
---|
246 | } |
---|
247 | else |
---|
248 | { |
---|
249 | printf("\nDecision:\n"); |
---|
250 | printf("\t%s (default class)\n", ClassName[DefaultClass]); |
---|
251 | } |
---|
252 | |
---|
253 | /* Prompt for what to do next */ |
---|
254 | |
---|
255 | while ( true ) |
---|
256 | { |
---|
257 | printf("\nRetry, new case or quit [r,n,q]: "); |
---|
258 | Reply = getchar(); |
---|
259 | SkipLine(Reply); |
---|
260 | switch ( Reply ) |
---|
261 | { |
---|
262 | case 'r': return; |
---|
263 | case 'n': Clear(); return; |
---|
264 | case 'q': exit(0); |
---|
265 | default: printf("Please enter 'r', 'n' or 'q'"); |
---|
266 | } |
---|
267 | } |
---|
268 | } |
---|
269 | |
---|
270 | |
---|
271 | |
---|
272 | /*************************************************************************/ |
---|
273 | /* */ |
---|
274 | /* Main routine for classifying items using a set of rules */ |
---|
275 | /* */ |
---|
276 | /*************************************************************************/ |
---|
277 | |
---|
278 | main(Argc, Argv) |
---|
279 | /* ---- */ |
---|
280 | int Argc; |
---|
281 | char *Argv[]; |
---|
282 | { |
---|
283 | int o; |
---|
284 | extern char *optarg; |
---|
285 | extern int optind; |
---|
286 | Attribute a; |
---|
287 | RuleNo r; |
---|
288 | |
---|
289 | PrintHeader("production rule interpreter"); |
---|
290 | |
---|
291 | /* Process options */ |
---|
292 | |
---|
293 | while ( (o = getopt(Argc, Argv, "tvf:")) != EOF ) |
---|
294 | { |
---|
295 | switch (o) |
---|
296 | { |
---|
297 | case 't': TRACE = 1; |
---|
298 | break; |
---|
299 | case 'v': VERBOSITY = 1; |
---|
300 | break; |
---|
301 | case 'f': FileName = optarg; |
---|
302 | break; |
---|
303 | case '?': printf("unrecognised option\n"); |
---|
304 | exit(1); |
---|
305 | } |
---|
306 | } |
---|
307 | |
---|
308 | /* Initialise */ |
---|
309 | |
---|
310 | GetNames(); |
---|
311 | GetRules(); |
---|
312 | |
---|
313 | if ( TRACE ) |
---|
314 | { |
---|
315 | ForEach(r, 1, NRules) |
---|
316 | { |
---|
317 | PrintRule(r); |
---|
318 | } |
---|
319 | printf("\nDefault class: %s\n", ClassName[DefaultClass]); |
---|
320 | } |
---|
321 | |
---|
322 | /* Allocate value ranges */ |
---|
323 | |
---|
324 | RangeDesc = (struct ValRange *) calloc(MaxAtt+1, sizeof(struct ValRange)); |
---|
325 | |
---|
326 | ForEach(a, 0, MaxAtt) |
---|
327 | { |
---|
328 | if ( MaxAttVal[a] ) |
---|
329 | { |
---|
330 | RangeDesc[a].Probability = |
---|
331 | (float *) calloc(MaxAttVal[a]+1, sizeof(float)); |
---|
332 | } |
---|
333 | } |
---|
334 | |
---|
335 | /* Consult */ |
---|
336 | |
---|
337 | Clear(); |
---|
338 | while ( true ) |
---|
339 | { |
---|
340 | InterpretRuleset(); |
---|
341 | } |
---|
342 | } |
---|