1 /***********************************************************************/
2 /* */
3 /* svm_classify.c */
4 /* */
5 /* Classification module of Support Vector Machine. */
6 /* */
7 /* Author: Thorsten Joachims */
8 /* Date: 02.07.02 */
9 /* */
10 /* Copyright (c) 2002 Thorsten Joachims - All rights reserved */
11 /* */
12 /* This software is available for non-commercial use only. It must */
13 /* not be modified and distributed without prior permission of the */
14 /* author. The author is not responsible for implications from the */
15 /* use of this software. */
16 /* */
17 /************************************************************************/
18
19 # include "svm_common.h"
20
21 char docfile[200];
22 char modelfile[200];
23 char predictionsfile[200];
24
25 void read_input_parameters(int, char **, char *, char *, char *, long *,
26 long *);
27 void print_help(void);
28
29
main(int argc,char * argv[])30 int main (int argc, char* argv[])
31 {
32 DOC *doc; /* test example */
33 WORD *words;
34 long max_docs,max_words_doc,lld;
35 long totdoc=0,queryid,slackid;
36 long correct=0,incorrect=0,no_accuracy=0;
37 long res_a=0,res_b=0,res_c=0,res_d=0,wnum,pred_format;
38 long j;
39 double t1,runtime=0;
40 double dist,doc_label,costfactor;
41 char *line,*comment;
42 FILE *predfl,*docfl;
43 MODEL *model;
44
45 read_input_parameters(argc,argv,docfile,modelfile,predictionsfile,
46 &verbosity,&pred_format);
47
48 nol_ll(docfile,&max_docs,&max_words_doc,&lld); /* scan size of input file */
49 max_words_doc+=2;
50 lld+=2;
51
52 line = (char *)my_malloc(sizeof(char)*lld);
53 words = (WORD *)my_malloc(sizeof(WORD)*(max_words_doc+10));
54
55 model=read_model(modelfile);
56
57 if(model->kernel_parm.kernel_type == 0) { /* linear kernel */
58 /* compute weight vector */
59 add_weight_vector_to_linear_model(model);
60 }
61
62 if(verbosity>=2) {
63 printf("Classifying test examples.."); fflush(stdout);
64 }
65
66 if ((docfl = fopen (docfile, "r")) == NULL)
67 { perror (docfile); exit (1); }
68 if ((predfl = fopen (predictionsfile, "w")) == NULL)
69 { perror (predictionsfile); exit (1); }
70
71 while((!feof(docfl)) && fgets(line,(int)lld,docfl)) {
72 if(line[0] == '#') continue; /* line contains comments */
73 parse_document(line,words,&doc_label,&queryid,&slackid,&costfactor,&wnum,
74 max_words_doc,&comment);
75 totdoc++;
76 if(model->kernel_parm.kernel_type == 0) { /* linear kernel */
77 for(j=0;(words[j]).wnum != 0;j++) { /* Check if feature numbers */
78 if((words[j]).wnum>model->totwords) /* are not larger than in */
79 (words[j]).wnum=0; /* model. Remove feature if */
80 } /* necessary. */
81 doc = create_example(-1,0,0,0.0,create_svector(words,comment,1.0));
82 t1=get_runtime();
83 dist=classify_example_linear(model,doc);
84 runtime+=(get_runtime()-t1);
85 free_example(doc,1);
86 }
87 else { /* non-linear kernel */
88 doc = create_example(-1,0,0,0.0,create_svector(words,comment,1.0));
89 t1=get_runtime();
90 dist=classify_example(model,doc);
91 runtime+=(get_runtime()-t1);
92 free_example(doc,1);
93 }
94 if(dist>0) {
95 if(pred_format==0) { /* old weired output format */
96 fprintf(predfl,"%.8g:+1 %.8g:-1\n",dist,-dist);
97 }
98 if(doc_label>0) correct++; else incorrect++;
99 if(doc_label>0) res_a++; else res_b++;
100 }
101 else {
102 if(pred_format==0) { /* old weired output format */
103 fprintf(predfl,"%.8g:-1 %.8g:+1\n",-dist,dist);
104 }
105 if(doc_label<0) correct++; else incorrect++;
106 if(doc_label>0) res_c++; else res_d++;
107 }
108 if(pred_format==1) { /* output the value of decision function */
109 fprintf(predfl,"%.8g\n",dist);
110 }
111 if((int)(0.01+(doc_label*doc_label)) != 1)
112 { no_accuracy=1; } /* test data is not binary labeled */
113 if(verbosity>=2) {
114 if(totdoc % 100 == 0) {
115 printf("%ld..",totdoc); fflush(stdout);
116 }
117 }
118 }
119 free(line);
120 free(words);
121 free_model(model,1);
122
123 if(verbosity>=2) {
124 printf("done\n");
125
126 /* Note by Gary Boone Date: 29 April 2000 */
127 /* o Timing is inaccurate. The timer has 0.01 second resolution. */
128 /* Because classification of a single vector takes less than */
129 /* 0.01 secs, the timer was underflowing. */
130 printf("Runtime (without IO) in cpu-seconds: %.2f\n",
131 (float)(runtime/100.0));
132
133 }
134 if((!no_accuracy) && (verbosity>=1)) {
135 printf("Accuracy on test set: %.2f%% (%ld correct, %ld incorrect, %ld total)\n",(float)(correct)*100.0/totdoc,correct,incorrect,totdoc);
136 printf("Precision/recall on test set: %.2f%%/%.2f%%\n",(float)(res_a)*100.0/(res_a+res_b),(float)(res_a)*100.0/(res_a+res_c));
137 }
138
139 return(0);
140 }
141
read_input_parameters(int argc,char ** argv,char * docfile,char * modelfile,char * predictionsfile,long int * verbosity,long int * pred_format)142 void read_input_parameters(int argc, char **argv, char *docfile,
143 char *modelfile, char *predictionsfile,
144 long int *verbosity, long int *pred_format)
145 {
146 long i;
147
148 /* set default */
149 strcpy (modelfile, "svm_model");
150 strcpy (predictionsfile, "svm_predictions");
151 (*verbosity)=2;
152 (*pred_format)=1;
153
154 for(i=1;(i<argc) && ((argv[i])[0] == '-');i++) {
155 switch ((argv[i])[1])
156 {
157 case 'h': print_help(); exit(0);
158 case 'v': i++; (*verbosity)=atol(argv[i]); break;
159 case 'f': i++; (*pred_format)=atol(argv[i]); break;
160 default: printf("\nUnrecognized option %s!\n\n",argv[i]);
161 print_help();
162 exit(0);
163 }
164 }
165 if((i+1)>=argc) {
166 printf("\nNot enough input parameters!\n\n");
167 print_help();
168 exit(0);
169 }
170 strcpy (docfile, argv[i]);
171 strcpy (modelfile, argv[i+1]);
172 if((i+2)<argc) {
173 strcpy (predictionsfile, argv[i+2]);
174 }
175 if(((*pred_format) != 0) && ((*pred_format) != 1)) {
176 printf("\nOutput format can only take the values 0 or 1!\n\n");
177 print_help();
178 exit(0);
179 }
180 }
181
print_help(void)182 void print_help(void)
183 {
184 printf("\nSVM-light %s: Support Vector Machine, classification module %s\n",VERSION_SVMLIGHT, VERSION_DATE_SVMLIGHT);
185 copyright_notice();
186 printf(" usage: svm_classify [options] example_file model_file output_file\n\n");
187 printf("options: -h -> this help\n");
188 printf(" -v [0..3] -> verbosity level (default 2)\n");
189 printf(" -f [0,1] -> 0: old output format of V1.0\n");
190 printf(" -> 1: output the value of decision function (default)\n\n");
191 }
192
193
194
195
196