| 89 | |
| 90 | |
| 91 | custom_decision_function train ( |
| 92 | const std::vector<sample_type>& samples, |
| 93 | const std::vector<double>& labels |
| 94 | ) const |
| 95 | /*! |
| 96 | requires |
| 97 | - is_binary_classification_problem(samples, labels) == true |
| 98 | (e.g. labels consists of only +1 and -1 values, samples.size() == labels.size()) |
| 99 | ensures |
| 100 | - returns a decision function F with the following properties: |
| 101 | - if (new_x is a sample predicted have +1 label) then |
| 102 | - F(new_x) >= 0 |
| 103 | - else |
| 104 | - F(new_x) < 0 |
| 105 | !*/ |
| 106 | { |
| 107 | sample_type positive_center, negative_center; |
| 108 | |
| 109 | // compute sums of each class |
| 110 | positive_center = 0; |
| 111 | negative_center = 0; |
| 112 | for (unsigned long i = 0; i < samples.size(); ++i) |
| 113 | { |
| 114 | if (labels[i] == +1) |
| 115 | positive_center += samples[i]; |
| 116 | else // this is a -1 sample |
| 117 | negative_center += samples[i]; |
| 118 | } |
| 119 | |
| 120 | // divide by number of +1 samples |
| 121 | positive_center /= sum(mat(labels) == +1); |
| 122 | // divide by number of -1 samples |
| 123 | negative_center /= sum(mat(labels) == -1); |
| 124 | |
| 125 | custom_decision_function df; |
| 126 | df.positive_center = positive_center; |
| 127 | df.negative_center = negative_center; |
| 128 | |
| 129 | return df; |
| 130 | } |
| 131 | }; |
| 132 | |
| 133 | // ---------------------------------------------------------------------------------------- |