[1806] | 1 | /*
|
---|
| 2 | * SVM.NET Library
|
---|
| 3 | * Copyright (C) 2008 Matthew Johnson
|
---|
| 4 | *
|
---|
| 5 | * This program is free software: you can redistribute it and/or modify
|
---|
| 6 | * it under the terms of the GNU General Public License as published by
|
---|
| 7 | * the Free Software Foundation, either version 3 of the License, or
|
---|
| 8 | * (at your option) any later version.
|
---|
| 9 | *
|
---|
| 10 | * This program is distributed in the hope that it will be useful,
|
---|
| 11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
| 12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
---|
| 13 | * GNU General Public License for more details.
|
---|
| 14 | *
|
---|
| 15 | * You should have received a copy of the GNU General Public License
|
---|
| 16 | * along with this program. If not, see <http://www.gnu.org/licenses/>.
|
---|
| 17 | */
|
---|
| 18 |
|
---|
| 19 |
|
---|
| 20 | using System;
|
---|
| 21 | using System.IO;
|
---|
| 22 | using System.Diagnostics;
|
---|
| 23 |
|
---|
| 24 | namespace SVM
|
---|
| 25 | {
|
---|
| 26 | /// <remarks>
|
---|
| 27 | /// Class containing the routines to perform class membership prediction using a trained SVM.
|
---|
| 28 | /// </remarks>
|
---|
| 29 | public static class Prediction
|
---|
| 30 | {
|
---|
| 31 | /// <summary>
|
---|
| 32 | /// Predicts the class memberships of all the vectors in the problem.
|
---|
| 33 | /// </summary>
|
---|
| 34 | /// <param name="problem">The SVM Problem to solve</param>
|
---|
| 35 | /// <param name="outputFile">File for result output</param>
|
---|
| 36 | /// <param name="model">The Model to use</param>
|
---|
| 37 | /// <param name="predict_probability">Whether to output a distribution over the classes</param>
|
---|
| 38 | /// <returns>Percentage correctly labelled</returns>
|
---|
| 39 | public static double Predict(
|
---|
| 40 | Problem problem,
|
---|
| 41 | string outputFile,
|
---|
| 42 | Model model,
|
---|
| 43 | bool predict_probability)
|
---|
| 44 | {
|
---|
| 45 | int correct = 0;
|
---|
| 46 | int total = 0;
|
---|
| 47 | double error = 0;
|
---|
| 48 | double sumv = 0, sumy = 0, sumvv = 0, sumyy = 0, sumvy = 0;
|
---|
| 49 | StreamWriter output = outputFile != null ? new StreamWriter(outputFile) : null;
|
---|
| 50 |
|
---|
| 51 | SvmType svm_type = Procedures.svm_get_svm_type(model);
|
---|
| 52 | int nr_class = Procedures.svm_get_nr_class(model);
|
---|
| 53 | int[] labels = new int[nr_class];
|
---|
| 54 | double[] prob_estimates = null;
|
---|
| 55 |
|
---|
| 56 | if (predict_probability)
|
---|
| 57 | {
|
---|
| 58 | if (svm_type == SvmType.EPSILON_SVR || svm_type == SvmType.NU_SVR)
|
---|
| 59 | {
|
---|
| 60 | Console.WriteLine("Prob. model for test data: target value = predicted value + z,\nz: Laplace distribution e^(-|z|/sigma)/(2sigma),sigma=" + Procedures.svm_get_svr_probability(model));
|
---|
| 61 | }
|
---|
| 62 | else
|
---|
| 63 | {
|
---|
| 64 | Procedures.svm_get_labels(model, labels);
|
---|
| 65 | prob_estimates = new double[nr_class];
|
---|
| 66 | if (output != null)
|
---|
| 67 | {
|
---|
| 68 | output.Write("labels");
|
---|
| 69 | for (int j = 0; j < nr_class; j++)
|
---|
| 70 | {
|
---|
| 71 | output.Write(" " + labels[j]);
|
---|
| 72 | }
|
---|
| 73 | output.Write("\n");
|
---|
| 74 | }
|
---|
| 75 | }
|
---|
| 76 | }
|
---|
| 77 | for (int i = 0; i < problem.Count; i++)
|
---|
| 78 | {
|
---|
| 79 | double target = problem.Y[i];
|
---|
| 80 | Node[] x = problem.X[i];
|
---|
| 81 |
|
---|
| 82 | double v;
|
---|
| 83 | if (predict_probability && (svm_type == SvmType.C_SVC || svm_type == SvmType.NU_SVC))
|
---|
| 84 | {
|
---|
| 85 | v = Procedures.svm_predict_probability(model, x, prob_estimates);
|
---|
| 86 | if (output != null)
|
---|
| 87 | {
|
---|
| 88 | output.Write(v + " ");
|
---|
| 89 | for (int j = 0; j < nr_class; j++)
|
---|
| 90 | {
|
---|
| 91 | output.Write(prob_estimates[j] + " ");
|
---|
| 92 | }
|
---|
| 93 | output.Write("\n");
|
---|
| 94 | }
|
---|
| 95 | }
|
---|
| 96 | else
|
---|
| 97 | {
|
---|
| 98 | v = Procedures.svm_predict(model, x);
|
---|
| 99 | if(output != null)
|
---|
| 100 | output.Write(v + "\n");
|
---|
| 101 | }
|
---|
| 102 |
|
---|
| 103 | if (v == target)
|
---|
| 104 | ++correct;
|
---|
| 105 | error += (v - target) * (v - target);
|
---|
| 106 | sumv += v;
|
---|
| 107 | sumy += target;
|
---|
| 108 | sumvv += v * v;
|
---|
| 109 | sumyy += target * target;
|
---|
| 110 | sumvy += v * target;
|
---|
| 111 | ++total;
|
---|
| 112 | }
|
---|
| 113 | if(output != null)
|
---|
| 114 | output.Close();
|
---|
| 115 | return (double)correct / total;
|
---|
| 116 | }
|
---|
| 117 |
|
---|
| 118 | /// <summary>
|
---|
| 119 | /// Predict the class for a single input vector.
|
---|
| 120 | /// </summary>
|
---|
| 121 | /// <param name="model">The Model to use for prediction</param>
|
---|
| 122 | /// <param name="x">The vector for which to predict class</param>
|
---|
| 123 | /// <returns>The result</returns>
|
---|
| 124 | public static double Predict(Model model, Node[] x)
|
---|
| 125 | {
|
---|
| 126 | return Procedures.svm_predict(model, x);
|
---|
| 127 | }
|
---|
| 128 |
|
---|
| 129 | /// <summary>
|
---|
| 130 | /// Predicts a class distribution for the single input vector.
|
---|
| 131 | /// </summary>
|
---|
| 132 | /// <param name="model">Model to use for prediction</param>
|
---|
| 133 | /// <param name="x">The vector for which to predict the class distribution</param>
|
---|
| 134 | /// <returns>A probability distribtion over classes</returns>
|
---|
| 135 | public static double[] PredictProbability(Model model, Node[] x)
|
---|
| 136 | {
|
---|
| 137 | SvmType svm_type = Procedures.svm_get_svm_type(model);
|
---|
| 138 | if (svm_type != SvmType.C_SVC && svm_type != SvmType.NU_SVC)
|
---|
| 139 | throw new Exception("Model type " + svm_type + " unable to predict probabilities.");
|
---|
| 140 | int nr_class = Procedures.svm_get_nr_class(model);
|
---|
| 141 | double[] probEstimates = new double[nr_class];
|
---|
| 142 | Procedures.svm_predict_probability(model, x, probEstimates);
|
---|
| 143 | return probEstimates;
|
---|
| 144 | }
|
---|
| 145 |
|
---|
| 146 | private static void exit_with_help()
|
---|
| 147 | {
|
---|
| 148 | Debug.Write("usage: svm_predict [options] test_file model_file output_file\n" + "options:\n" + "-b probability_estimates: whether to predict probability estimates, 0 or 1 (default 0); one-class SVM not supported yet\n");
|
---|
| 149 | Environment.Exit(1);
|
---|
| 150 | }
|
---|
| 151 |
|
---|
| 152 | /// <summary>
|
---|
| 153 | /// Legacy method, provided to allow usage as though this were the command line version of libsvm.
|
---|
| 154 | /// </summary>
|
---|
| 155 | /// <param name="args">Standard arguments passed to the svm_predict exectutable. See libsvm documentation for details.</param>
|
---|
| 156 | [Obsolete("Use the other version of Predict() instead")]
|
---|
| 157 | public static void Predict(params string[] args)
|
---|
| 158 | {
|
---|
| 159 | int i = 0;
|
---|
| 160 | bool predictProbability = false;
|
---|
| 161 |
|
---|
| 162 | // parse options
|
---|
| 163 | for (i = 0; i < args.Length; i++)
|
---|
| 164 | {
|
---|
| 165 | if (args[i][0] != '-')
|
---|
| 166 | break;
|
---|
| 167 | ++i;
|
---|
| 168 | switch (args[i - 1][1])
|
---|
| 169 | {
|
---|
| 170 |
|
---|
| 171 | case 'b':
|
---|
| 172 | predictProbability = int.Parse(args[i]) == 1;
|
---|
| 173 | break;
|
---|
| 174 |
|
---|
| 175 | default:
|
---|
| 176 | throw new ArgumentException("Unknown option");
|
---|
| 177 |
|
---|
| 178 | }
|
---|
| 179 | }
|
---|
| 180 | if (i >= args.Length)
|
---|
| 181 | throw new ArgumentException("No input, model and output files provided");
|
---|
| 182 |
|
---|
| 183 | Problem problem = Problem.Read(args[i]);
|
---|
| 184 | Model model = Model.Read(args[i + 1]);
|
---|
| 185 | Predict(problem, args[i + 2], model, predictProbability);
|
---|
| 186 | }
|
---|
| 187 | }
|
---|
| 188 | } |
---|