Go to the documentation of this file.00001 import libsvm.*;
00002 import java.io.*;
00003 import java.util.*;
00004
00005 class svm_train {
00006 private svm_parameter param;
00007 private svm_problem prob;
00008 private svm_model model;
00009 private String input_file_name;
00010 private String model_file_name;
00011 private String error_msg;
00012 private int cross_validation;
00013 private int nr_fold;
00014
00015 private static svm_print_interface svm_print_null = new svm_print_interface()
00016 {
00017 public void print(String s) {}
00018 };
00019
00020 private static void exit_with_help()
00021 {
00022 System.out.print(
00023 "Usage: svm_train [options] training_set_file [model_file]\n"
00024 +"options:\n"
00025 +"-s svm_type : set type of SVM (default 0)\n"
00026 +" 0 -- C-SVC\n"
00027 +" 1 -- nu-SVC\n"
00028 +" 2 -- one-class SVM\n"
00029 +" 3 -- epsilon-SVR\n"
00030 +" 4 -- nu-SVR\n"
00031 +"-t kernel_type : set type of kernel function (default 2)\n"
00032 +" 0 -- linear: u'*v\n"
00033 +" 1 -- polynomial: (gamma*u'*v + coef0)^degree\n"
00034 +" 2 -- radial basis function: exp(-gamma*|u-v|^2)\n"
00035 +" 3 -- sigmoid: tanh(gamma*u'*v + coef0)\n"
00036 +" 4 -- precomputed kernel (kernel values in training_set_file)\n"
00037 +"-d degree : set degree in kernel function (default 3)\n"
00038 +"-g gamma : set gamma in kernel function (default 1/num_features)\n"
00039 +"-r coef0 : set coef0 in kernel function (default 0)\n"
00040 +"-c cost : set the parameter C of C-SVC, epsilon-SVR, and nu-SVR (default 1)\n"
00041 +"-n nu : set the parameter nu of nu-SVC, one-class SVM, and nu-SVR (default 0.5)\n"
00042 +"-p epsilon : set the epsilon in loss function of epsilon-SVR (default 0.1)\n"
00043 +"-m cachesize : set cache memory size in MB (default 100)\n"
00044 +"-e epsilon : set tolerance of termination criterion (default 0.001)\n"
00045 +"-h shrinking : whether to use the shrinking heuristics, 0 or 1 (default 1)\n"
00046 +"-b probability_estimates : whether to train a SVC or SVR model for probability estimates, 0 or 1 (default 0)\n"
00047 +"-wi weight : set the parameter C of class i to weight*C, for C-SVC (default 1)\n"
00048 +"-v n : n-fold cross validation mode\n"
00049 +"-q : quiet mode (no outputs)\n"
00050 );
00051 System.exit(1);
00052 }
00053
00054 private void do_cross_validation()
00055 {
00056 int i;
00057 int total_correct = 0;
00058 double total_error = 0;
00059 double sumv = 0, sumy = 0, sumvv = 0, sumyy = 0, sumvy = 0;
00060 double[] target = new double[prob.l];
00061
00062 svm.svm_cross_validation(prob,param,nr_fold,target);
00063 if(param.svm_type == svm_parameter.EPSILON_SVR ||
00064 param.svm_type == svm_parameter.NU_SVR)
00065 {
00066 for(i=0;i<prob.l;i++)
00067 {
00068 double y = prob.y[i];
00069 double v = target[i];
00070 total_error += (v-y)*(v-y);
00071 sumv += v;
00072 sumy += y;
00073 sumvv += v*v;
00074 sumyy += y*y;
00075 sumvy += v*y;
00076 }
00077 System.out.print("Cross Validation Mean squared error = "+total_error/prob.l+"\n");
00078 System.out.print("Cross Validation Squared correlation coefficient = "+
00079 ((prob.l*sumvy-sumv*sumy)*(prob.l*sumvy-sumv*sumy))/
00080 ((prob.l*sumvv-sumv*sumv)*(prob.l*sumyy-sumy*sumy))+"\n"
00081 );
00082 }
00083 else
00084 {
00085 for(i=0;i<prob.l;i++)
00086 if(target[i] == prob.y[i])
00087 ++total_correct;
00088 System.out.print("Cross Validation Accuracy = "+100.0*total_correct/prob.l+"%\n");
00089 }
00090 }
00091
00092 private void run(String argv[]) throws IOException
00093 {
00094 parse_command_line(argv);
00095 read_problem();
00096 error_msg = svm.svm_check_parameter(prob,param);
00097
00098 if(error_msg != null)
00099 {
00100 System.err.print("Error: "+error_msg+"\n");
00101 System.exit(1);
00102 }
00103
00104 if(cross_validation != 0)
00105 {
00106 do_cross_validation();
00107 }
00108 else
00109 {
00110 model = svm.svm_train(prob,param);
00111 svm.svm_save_model(model_file_name,model);
00112 }
00113 }
00114
00115 public static void main(String argv[]) throws IOException
00116 {
00117 svm_train t = new svm_train();
00118 t.run(argv);
00119 }
00120
00121 private static double atof(String s)
00122 {
00123 double d = Double.valueOf(s).doubleValue();
00124 if (Double.isNaN(d) || Double.isInfinite(d))
00125 {
00126 System.err.print("NaN or Infinity in input\n");
00127 System.exit(1);
00128 }
00129 return(d);
00130 }
00131
00132 private static int atoi(String s)
00133 {
00134 return Integer.parseInt(s);
00135 }
00136
00137 private void parse_command_line(String argv[])
00138 {
00139 int i;
00140 svm_print_interface print_func = null;
00141
00142 param = new svm_parameter();
00143
00144 param.svm_type = svm_parameter.C_SVC;
00145 param.kernel_type = svm_parameter.RBF;
00146 param.degree = 3;
00147 param.gamma = 0;
00148 param.coef0 = 0;
00149 param.nu = 0.5;
00150 param.cache_size = 100;
00151 param.C = 1;
00152 param.eps = 1e-3;
00153 param.p = 0.1;
00154 param.shrinking = 1;
00155 param.probability = 0;
00156 param.nr_weight = 0;
00157 param.weight_label = new int[0];
00158 param.weight = new double[0];
00159 cross_validation = 0;
00160
00161
00162 for(i=0;i<argv.length;i++)
00163 {
00164 if(argv[i].charAt(0) != '-') break;
00165 if(++i>=argv.length)
00166 exit_with_help();
00167 switch(argv[i-1].charAt(1))
00168 {
00169 case 's':
00170 param.svm_type = atoi(argv[i]);
00171 break;
00172 case 't':
00173 param.kernel_type = atoi(argv[i]);
00174 break;
00175 case 'd':
00176 param.degree = atoi(argv[i]);
00177 break;
00178 case 'g':
00179 param.gamma = atof(argv[i]);
00180 break;
00181 case 'r':
00182 param.coef0 = atof(argv[i]);
00183 break;
00184 case 'n':
00185 param.nu = atof(argv[i]);
00186 break;
00187 case 'm':
00188 param.cache_size = atof(argv[i]);
00189 break;
00190 case 'c':
00191 param.C = atof(argv[i]);
00192 break;
00193 case 'e':
00194 param.eps = atof(argv[i]);
00195 break;
00196 case 'p':
00197 param.p = atof(argv[i]);
00198 break;
00199 case 'h':
00200 param.shrinking = atoi(argv[i]);
00201 break;
00202 case 'b':
00203 param.probability = atoi(argv[i]);
00204 break;
00205 case 'q':
00206 print_func = svm_print_null;
00207 i--;
00208 break;
00209 case 'v':
00210 cross_validation = 1;
00211 nr_fold = atoi(argv[i]);
00212 if(nr_fold < 2)
00213 {
00214 System.err.print("n-fold cross validation: n must >= 2\n");
00215 exit_with_help();
00216 }
00217 break;
00218 case 'w':
00219 ++param.nr_weight;
00220 {
00221 int[] old = param.weight_label;
00222 param.weight_label = new int[param.nr_weight];
00223 System.arraycopy(old,0,param.weight_label,0,param.nr_weight-1);
00224 }
00225
00226 {
00227 double[] old = param.weight;
00228 param.weight = new double[param.nr_weight];
00229 System.arraycopy(old,0,param.weight,0,param.nr_weight-1);
00230 }
00231
00232 param.weight_label[param.nr_weight-1] = atoi(argv[i-1].substring(2));
00233 param.weight[param.nr_weight-1] = atof(argv[i]);
00234 break;
00235 default:
00236 System.err.print("Unknown option: " + argv[i-1] + "\n");
00237 exit_with_help();
00238 }
00239 }
00240
00241 svm.svm_set_print_string_function(print_func);
00242
00243
00244
00245 if(i>=argv.length)
00246 exit_with_help();
00247
00248 input_file_name = argv[i];
00249
00250 if(i<argv.length-1)
00251 model_file_name = argv[i+1];
00252 else
00253 {
00254 int p = argv[i].lastIndexOf('/');
00255 ++p;
00256 model_file_name = argv[i].substring(p)+".model";
00257 }
00258 }
00259
00260
00261
00262 private void read_problem() throws IOException
00263 {
00264 BufferedReader fp = new BufferedReader(new FileReader(input_file_name));
00265 Vector<Double> vy = new Vector<Double>();
00266 Vector<svm_node[]> vx = new Vector<svm_node[]>();
00267 int max_index = 0;
00268
00269 while(true)
00270 {
00271 String line = fp.readLine();
00272 if(line == null) break;
00273
00274 StringTokenizer st = new StringTokenizer(line," \t\n\r\f:");
00275
00276 vy.addElement(atof(st.nextToken()));
00277 int m = st.countTokens()/2;
00278 svm_node[] x = new svm_node[m];
00279 for(int j=0;j<m;j++)
00280 {
00281 x[j] = new svm_node();
00282 x[j].index = atoi(st.nextToken());
00283 x[j].value = atof(st.nextToken());
00284 }
00285 if(m>0) max_index = Math.max(max_index, x[m-1].index);
00286 vx.addElement(x);
00287 }
00288
00289 prob = new svm_problem();
00290 prob.l = vy.size();
00291 prob.x = new svm_node[prob.l][];
00292 for(int i=0;i<prob.l;i++)
00293 prob.x[i] = vx.elementAt(i);
00294 prob.y = new double[prob.l];
00295 for(int i=0;i<prob.l;i++)
00296 prob.y[i] = vy.elementAt(i);
00297
00298 if(param.gamma == 0 && max_index > 0)
00299 param.gamma = 1.0/max_index;
00300
00301 if(param.kernel_type == svm_parameter.PRECOMPUTED)
00302 for(int i=0;i<prob.l;i++)
00303 {
00304 if (prob.x[i][0].index != 0)
00305 {
00306 System.err.print("Wrong kernel matrix: first column must be 0:sample_serial_number\n");
00307 System.exit(1);
00308 }
00309 if ((int)prob.x[i][0].value <= 0 || (int)prob.x[i][0].value > max_index)
00310 {
00311 System.err.print("Wrong input format: sample_serial_number out of range\n");
00312 System.exit(1);
00313 }
00314 }
00315
00316 fp.close();
00317 }
00318 }